A review of uncertainty visualization errors:

Working memory as an explanatory theory - eBook

PDF pdf download

https://ebooksecure.com/download/a-review-of-uncertainty-
visualization-errors-working-memory-as-an-explanatory-theory-
ebook-pdf/

Download more ebook from https://ebooksecure.com

We believe these products will be a great fit for you. Click
the link to download now, or visit ebooksecure.com
to discover even more!

Cultivating an understanding of curiosity as a seed for

creativity - eBook PDF

https://ebooksecure.com/download/cultivating-an-understanding-of-
curiosity-as-a-seed-for-creativity-ebook-pdf/

Translation as a Science Translation as an Art, 2e: A

Practical Approach 2nd Edition Noa Talaván - eBook PDF

https://ebooksecure.com/download/translation-as-a-science-
translation-as-an-art-2e-a-practical-approach-ebook-pdf/

(eBook PDF) Graduate Review of Tonal Theory: A

Recasting of Common-Practice Harmony, Form, and
Counterpoint

http://ebooksecure.com/product/ebook-pdf-graduate-review-of-
tonal-theory-a-recasting-of-common-practice-harmony-form-and-
counterpoint/

(eBook PDF) A Guide to Writing as an Engineer, 4th

Edition

http://ebooksecure.com/product/ebook-pdf-a-guide-to-writing-as-
an-engineer-4th-edition/
(eBook PDF) Mechanisms of Memory 2nd Edition

http://ebooksecure.com/product/ebook-pdf-mechanisms-of-
memory-2nd-edition/

(eBook PDF) Student Workbook to Accompany Graduate

Review of Tonal Theory

http://ebooksecure.com/product/ebook-pdf-student-workbook-to-
accompany-graduate-review-of-tonal-theory/

Dimensions of Uncertainty in Communication Engineering

1st Editon- eBook PDF

https://ebooksecure.com/download/dimensions-of-uncertainty-in-
communication-engineering-ebook-pdf/

(eBook PDF) Probability and Statistics: The Science of

Uncertainty Second Edition

http://ebooksecure.com/product/ebook-pdf-probability-and-
statistics-the-science-of-uncertainty-second-edition/

Pediatric Endocrinology and Inborn Errors of Metabolism

2nd Edition Kyriakie Sarafoglou - eBook PDF

https://ebooksecure.com/download/pediatric-endocrinology-and-
inborn-errors-of-metabolism-ebook-pdf/
 ARTICLE IN PRESS

A review of uncertainty
visualization errors: Working
memory as an explanatory theory
Lace Padillaa,∗, Spencer C. Castrob, and Helia Hosseinpoura
a
Cognitive and Information Sciences Department, University of California Merced, Merced, CA, United States
b
Management of Complex Systems Department, University of California Merced, Merced, CA, United States
∗
Corresponding author: e-mail address: [email protected]

Contents
1. Introduction 2
2. Visualization decision-making framework 7
2.1 Visual array and attention 7
2.2 Working memory 8
2.3 Visual description 10
2.4 Graph schemas 11
2.5 Matching process 11
2.6 Instantiated graph schema 12
2.7 Message assembly 12
2.8 Conceptual question 12
2.9 Decision-making 13
2.10 Behavior 14
3. Uncertainty visualization errors 14
3.1 Early-stage processing errors 14
3.2 Middle-stage processing errors 22
3.3 Late-stage errors 31
4. Conclusions 34
References 36

Abstract
Uncertainty communicators often use visualizations to express the unknowns in data,
statistical analyses, and forecasts. Well-designed visualizations can clearly and effectively
convey uncertainty, which is vital for ensuring transparency, accuracy, and scientific
credibility. However, poorly designed uncertainty visualizations can lead to misunder-
standings of the underlying data and result in poor decision-making. In this chapter,
we present a discussion of errors in uncertainty visualization research and current
approaches to evaluation. Researchers consistently find that uncertainty visualizations
requiring mental operations, rather than judgments guided by the visual system, lead
to more errors. To summarize this work, we propose that increased working memory

Psychology of Learning and Motivation Copyright # 2021 Elsevier Inc. 1

ISSN 0079-7421 All rights reserved.
https://doi.org/10.1016/bs.plm.2021.03.001
 ARTICLE IN PRESS

2 Lace Padilla et al.

demand may account for many observed uncertainty visualization errors. In particular,
the most common uncertainty visualization in scientific communication (e.g., variants of
confidence intervals) produces systematic errors that may be attributable to the appli-
cation of working memory or lack thereof. To create a more effective uncertainty
visualization, we recommend that data communicators seek a sweet spot in the work-
ing memory required by various tasks and visualization users. Further, we also recom-
mend that more work be done to evaluate the working memory demand of uncertainty
visualizations and visualizations more broadly.

1. Introduction
From simple analyses, such as those used in introductory statistics text-
books, to the complex forecasts of pandemic projection models, uncertainty
presents a difficult challenge for those seeking to represent and interpret it.
Uncertainties that can arise throughout a modeling and analysis pipeline
(Pang, Wittenbrink, & Lodha, 1997) are of interest to many fields. To con-
strain the complex category of uncertainty to its component parts, scholars
commonly distinguish between several types of uncertainty: ontological
(uncertainty created by the accuracy of the subjectively described reality
depicted in the model), epistemic (limited knowledge producing uncertainty),
and aleatoric (inherent irreducible randomness of a process; Spiegelhalter,
2017). Additionally, quantified forms of aleatoric and epistemic uncertainty
are referred to as risk in decision-making domains (Knight, 2012). In this chap-
ter, we define uncertainty to encompass quantifiable and visualizable uncer-
tainty, such as a probability distribution.
Many people have difficulty reasoning with even simple forms of uncer-
tainty (Gal, 2002). One study found that 16–20% of 463 college-educated
participants could not correctly answer the question, “Which represents the
larger risk: 1%, 5%, or 10%?” (Lipkus, Samsa, & Rimer, 2001). Other work
finds that even experts with training in statistics commonly misunderstand
how to interpret statistical significance from frequentist 95% confidence inter-
vals (Belia, Fidler, Williams, & Cumming, 2005). These findings—that even
simple forms of uncertainty are challenging for college graduates and statisti-
cians to understand—should concern both the scientific community and soci-
ety. We should be concerned because we all make both small- and large-scale
decisions with uncertainty throughout our lives, such as picking stocks to
invest in or evaluating our pandemic risk.
In the context of textual expressions of uncertainty, researchers propose
that people have difficulty understanding probabilities when expressed as a
 ARTICLE IN PRESS

Uncertainty visualization errors 3

percent (e.g., 10% chance of rain), because this framing is not how we expe-
rience probabilities in our daily lives (Gigerenzer & Hoffrage, 1995). A sub-
stantial body of research demonstrates that if we express uncertainty in
the form of frequency (e.g., it will rain 1 of 10 times), the representation
becomes more intuitive (e.g., Gigerenzer, 1996, 2008; Gigerenzer &
Gaissmaier, 2011; Gigerenzer & Hoffrage, 1995; Gigerenzer, Todd, &
ABC Research Group, 2000; Hoffrage & Gigerenzer, 1998). This line of
inquiry takes the perspective that humans can effectively reason with uncer-
tainty if, and only if, the information is presented in an intuitive way.
In addition to research on textural expressions of uncertainty, a large body of
evidence demonstrates that communicating uncertainty visually can help peo-
ple make more effective judgments about risk (for reviews see, Kinkeldey,
MacEachren, Riveiro, & Schiewe, 2017; Kinkeldey, MacEachren, &
Schiewe, 2014; Maceachren et al., 2005; Padilla, Kay, & Hullman, 2021).
Researchers propose that visualizations leverage the substantial processing
power of the visual system (Zacks & Franconeri, 2020), recruiting roughly half
of the brain (Van Essen, Anderson, & Felleman, 1992). Visualizations allow a
viewer’s visual system to complete some complex processing efficiently, such as
pattern recognition and data comparisons (Szafir, Haroz, Gleicher, &
Franconeri, 2016), which would be more challenging to do mathematically.
The power and efficiency of the visual system creates an advantage for visual-
izations over textual expressions of uncertainty. For example, consider how
long it takes to read about the following two treatments and how challenging
it is to decide which is riskier.
Treatment A: 3 of 10 patients have side effects.
Treatment B: 6 of 45 patients have side effects.
Now consider the same comparison of treatments but visualized using
the icon array in Fig. 1.

Treatment A Treatment B

Fig. 1 Icon arrays showing the proportion of patients with side effects in red after
receiving hypothetical treatments A or B.
 ARTICLE IN PRESS

4 Lace Padilla et al.

The red icons in Treatment A represent a larger portion of side effects

than the red icons in Treatment B. Icon arrays afford visual comparisons
that are relatively quick and easy for the visual system to compute, using
Gestalt grouping principles, which we will discuss in the Early-Stage
Processing Errors section. The visual comparison process above does not nec-
essarily require any mathematical calculation. A viewer can arrive at the correct
answer, that treatment A is riskier than B, by visually comparing the propor-
tion of side effects for each treatment and determining that A is larger. The
viewer does not need to calculate the exact proportions to accomplish this
task. Researchers have extensively studied icon arrays in the context of health
care communication; they find that icon arrays consistently help people
understand probabilities of risk and can be easier to understand than textual
representations of probabilities (for reviews, see Fagerlin, Zikmund-Fisher,
& Ubel, 2011; Garcia-Retamero & Cokely, 2017; Waters, Fagerlin, &
Zikmund-Fisher, 2016).
Many researchers have demonstrated that visualizations of uncertainty can
lead to better judgments than textual descriptions of the same information
(Fagerlin, Wang, & Ubel, 2005; Feldman-Stewart, Brundage, & Zotov,
2007; Fernandes, Walls, Munson, Hullman, & Kay, 2018; Garcia-Retamero
& Galesic, 2009a, 2009b; Garcia-Retamero, Galesic, & Gigerenzer, 2010;
Garcia-Retamero, Okan, & Cokely, 2012; Hawley et al., 2008; Tait,
Voepel-Lewis, Zikmund-Fisher, & Fagerlin, 2010; Waters et al., 2016;
Waters, Weinstein, Colditz, & Emmons, 2006). For example, one study
presented a mixed group of older adults and students with probabilities
via text (e.g., “aspirin can reduce the risk of having a stroke or heart attack
by 13%”). A second group was shown textual and icon arrays of this infor-
mation (Galesic, Garcia-Retamero, & Gigerenzer, 2009). Participants were
asked to estimate the number of people out of 1000 who had a stroke if
they did and did not take aspirin. Participants who were provided with
the icon arrays in addition to the textual information made significantly
more accurate judgments. Researchers have also documented improve-
ments compared to text for more complex visualizations (e.g., Fernandes
et al., 2018) (for a review of effective uncertainty visualization technique,
see Padilla et al., 2021).
In addition to mounting evidence illustrating the utility of uncertainty
visualizations, a number of studies have also documented reasoning errors
(e.g., Belia et al., 2005; Correll & Gleicher, 2014; Joslyn & LeClerc,
2013; Padilla, Creem-Regehr, & Thompson, 2020; Padilla, Ruginski, &
Creem-Regehr, 2017; Ruginski et al., 2016). Errors due to interpreting
 ARTICLE IN PRESS

Uncertainty visualization errors 5

uncertainty visualizations can exacerbate the difficulty people have in rea-

soning with uncertainty.
An uncertainty visualization can unintentionally mislead its viewers,
which results in poorer decision-making (e.g., Broad, Leiserowitz,
Weinkle, & Steketee, 2007). For example, the Cone of Uncertainty—
produced by the National Hurricane Center—has become one of the most
notorious uncertainty visualizations (see, Fig. 2). The Cone of Uncertainty is
intended to show the forecasted path of a storm with the centerline rep-
resenting the mean prediction and the edge of the cone denoting a 66.6%
confidence interval around the mean. When people are not provided with
additional information about what the cone is intended to represent, they
believe that the cone shows the size of the hurricane growing over time
(Padilla et al., 2017). Instead, the cone is intended to show that the uncer-
tainty in the storm’s path increases with time from the initial forecast. The
concept of uncertainty increasing over time can be intuitive. For example,

Fig. 2 Example hurricane track forecast cone produced by National Hurricane Center
(https://www.nhc.noaa.gov/aboutcone.shtml).
 ARTICLE IN PRESS

6 Lace Padilla et al.

it is easier to predict the temperature for tomorrow than the temperature for
2 weeks from now. However, when uncertainty in the storm’s path is rep-
resented visually with a cone-like visualization, it requires effort to under-
stand it as anything other than the size of the storm.
Within traditional uncertainty visualization research, practitioners
commonly recommend a set of best practices or general principles without
positing cognitive theories as to why a visualization might produce errors.
However, uncertainty visualization researchers are increasingly interested
in cognitive perspectives (Fernandes et al., 2018; Hullman, Kay, Kim, &
Shrestha, 2017; Kale, Kay, & Hullman, 2020; Kale, Nguyen, Kay, &
Hullman, 2018; Kim, Walls, Krafft, & Hullman, 2019). Notably, Kim
et al. (2019) propose a Bayesian cognitive modeling approach to incorporate
prior beliefs and update evaluations of uncertainty visualizations. Also, Joslyn
and Savelli (2020) detail the cognitive mechanisms associated with a specific
type of reasoning error in uncertainty visualization. Although prior approaches
have detailed the cognitive aspects of reasoning with uncertainty visualizations,
they do not offer a unified theory that describes the sources of errors across
visualization types. As a result, accurately predicting when a new type of
uncertainty visualization will fall into the category of helpful or harmful is
difficult.
The current chapter seeks to bridge this gap in knowledge by providing
a unifying theory for why errors occur when making decisions with uncer-
tainty visualizations. We begin this work by describing a cognitive framework
for how decisions are made with visualizations (Padilla, Creem-Regehr,
Hegarty, & Stefanucci, 2018), which we subsequently use as a tool to ground
empirical work on errors in uncertainty visualization. Then, we review
behavioral evidence of using uncertainty visualizations with a focus on when
errors or misunderstandings occur, in order to find commonalities among
these errors.
As a preview, researchers consistently observe errors when a visualization
or task requires a viewer to perform a complex mental computation to accu-
rately interpret the visual information. We propose that a unifying cognitive
process that predicts these errors is increased working memory or cognitive
effort. This chapter reviews research on working memory demand in the
context of visualizations and how working memory as a mental process
can potentially explain many of the errors observed in uncertainty-
visualization use.
 ARTICLE IN PRESS

Uncertainty visualization errors 7

2. Visualization decision-making framework

Reasoning errors with uncertainty visualizations have the potential to
arise at various stages in the decision-making process. If we have a clear
understanding of where an error occurs in this process, we can more clearly
develop interventions to help make more effective decisions with uncer-
tainty visualizations.
For this chapter, we will utilize a cognitive model that describes decision-
making with visualizations proposed by Padilla et al. (2018) (see Fig. 3). The
Padilla et al. (2018) model integrates a dual-process theory of decision-making
and a modern understanding of visualization comprehension and learning.

2.1 Visual array and attention

The Padilla et al. (2018) model begins with a visual array, which is the
unprocessed neuronal firing in response to a stimulus. Bottom-up and
top-down attention guide a viewer’s gaze around the image. Bottom-up atten-
tion refers to how the visual system is guided to elements in the visualization
based on visual salience. Errors that occur from bottom-up attention result
from the visualization directing a viewer’s attention to task-irrelevant infor-
mation. Top-down attention is how the viewer controls his or her gaze
around a visualization. Top-down attention is based on the viewer’s goals,
experiences, and other individual differences. For example, Kim et al. (2019)
capture the influence of top-down attention on decision-making with

Working
Memory

Conceptual
Question

inference
top-down
attention Visual Instantiated
Visual Array Description Graph Schema
Conceptual
bottom-up Message decision Behavior
attention MATCH message
assembly making

processes
influences

Fig. 3 Visualization decision-making model proposed by Padilla et al. (2018).

Reproduced per CC-BY license from Padilla, L., Creem-Regehr, S., Hegarty, M., &
Stefanucci, J. (2018). Decision making with visualizations: A cognitive framework across dis-
ciplines. Cognitive Research: Principles and Implications, 3, 29.
 ARTICLE IN PRESS

8 Lace Padilla et al.

Bayesian priors representing previous beliefs and experiences. Errors that

arise due to top-down attention may come from users having an incomplete
understanding of how to achieve their goals, having experiences that tell
them to search through an image ineffectively, or other potential biases
based on long-term knowledge.

2.2 Working memory

Working memory (located in the circle at the top of Fig. 3) is a cognitive pro-
cess that can influence most visualization decision-making processes (Padilla
et al., 2018). The debate about how to define the term working memory is
ongoing, as it has differing characteristics in various fields (Cowan, 2017).
For this review, we will use the definition of working memory defined
by Cowan (2017), where working memory is a multi-component system.
Working memory maintains a finite amount of information for a short time
before that information is potentially stored in long-term memory. In the
context of uncertainty visualization, the term maintain means that when
viewers see a visualization, they store a mental representation of it in their
mind to update or manipulate later. For example, viewers might see a scatter
plot and want to find the data’s central tendency. In their mind, they would
mentally overlay a trend line onto their temporarily stored mental represen-
tation of the visualization. Within the traditional model of working mem-
ory, the visual-spatial sketch pad represents the mechanism that maintains
information from a visualization; a separate mechanism maintains phonolog-
ical information (Baddeley, 1992).
Working memory has a central executive that controls its multicomponent
functions, and it works to control attention while suppressing automatic
processes (Logie & Marchetti, 1991). For example, the process of explicitly
directing top-down attention requires working memory (Shipstead,
Harrison, & Engle, 2015), as in directing one’s attention away from salient
information in a visualization. An error occurs when bottom-up attention
guides the visual system to visually salient but task-irrelevant elements in
the visualization. The central executive exerts its control over the finite
amount of working memory available to simultaneously suppress
bottom-up attention shifting to task-irrelevant information and guide atten-
tion toward task-relevant stimuli.
Three types of errors may occur due to working memory relevant to
uncertainty visualizations: capacity limitation errors, failure to utilize work-
ing memory, and temporal decay errors. Researchers have traditionally
 ARTICLE IN PRESS

Uncertainty visualization errors 9

studied capacity limitation errors in the context of how many digits or items in
sequence participants can remember (Miller, 1956). More recent work sug-
gests that we tend to group information (e.g., chunk) rather than maintain the
information separately and that we can remember between three to five
chunks of information (Doumont, 2002). Errors may occur when viewing
uncertainty visualizations if a visualization requires the viewer to maintain
too much information in working memory, essentially surpassing the limited
working memory capacity. As a simple example, imagine a visualization that
maps elements of the data to color, opacity, texture, size, shape, and position.
To interpret the visualization correctly, one must maintain in working mem-
ory how each variable relates to the data. Working memory capacity may be
overloaded if people are asked to do a complex data analysis with such a
working-memory demanding visualization. Capacity limitation errors include
failing to integrate all of the relevant information in a visualization; not being
able to perform a mental computation on a visualization; or failing to main-
tain, switch, or update task goals.
The second category of errors related to working memory encompasses
viewers failing to use working memory when they should. By default, we
tend to make fast and automated decisions that use as little working memory
as possible (Type 1 processing) (Kahneman, 2011; Tversky & Kahneman,
1974). Type 1 processing is an adaptive strategy that we have developed
to minimize effort because effort is metabolically costly. Researchers esti-
mate that our brains account for 20–25% of our resting metabolism
(Leonard & Robertson, 1994). Voluntary effort may not exclusively account
for the mind’s propensity toward Type 1 processing, but a combination of
energy conservation and reserving limited capacity working memory vali-
dates the preference for fast and automated decisions (Kool & Botvinick,
2014). However, some visualizations require the use of working memory
to be understood correctly (Type 2 processing). For example, when viewing
the line chart in Fig. 4 that illustrates the impact of the Stand Your Ground
law on gun deaths in Florida, a viewer might not notice that the Y-axis is
inverted. Without using working memory, the viewer would assume that
the Stand Your Ground law correlated with a drop in gun deaths in
Florida. To interpret this visualization correctly, a viewer needs to activate
working memory to recognize that the Y-axis is inverted and reimagine the
data’s appropriate relationships.
The third type of error related to working memory results from forgetting
relevant information because working memory decays over time. For exam-
ple, if asked to memorize the sequence 9,875,341,890, recalling the numbers
 ARTICLE IN PRESS

10 Lace Padilla et al.

Fig. 4 Deceptive visualization showing the impact of the 2005 Stand Your Ground law
in Florida and the number of murders from firearms with the Y-axis reversed. This exam-
ple is based on a data visualization that was released to the public by Christine Chan at
Reuters (Pandey, Rall, Satterthwaite, Nov, & Bertini, 2015). Redrawn per CC-BY license from
Padilla, L., Creem-Regehr, S., Hegarty, M., & Stefanucci, J. (2018). Decision making with
visualizations: A cognitive framework across disciplines. Cognitive Research: Principles
and Implications, 3, 29.

after holding them in working memory for 5 s is easier than after 5 min. To
memorize such information and hold it in working memory for prolonged
periods, people generally chunk information, such as (987) 534–1890, and
then mentally rehearse the information. Without rehearsal, our ability to store
information begins to decay after approximately 5–10 s (Cowan, 2017). The
nature of the decay can vary due to the task, type of information, and indi-
vidual capacities (Cowan, Saults, & Nugent, 1997). Longer sequential visual-
ization tasks that require completion of longer-term goals may be error-prone
due to the degradation of working memory over time.

2.3 Visual description

The visual description (second box in Fig. 3) is the resultant mental conception of
the visualization’s information after top-down and bottom-up processing have
guided the extraction of information. Note that the visual description is not
identical to the visualization; its generation is dependent on what the viewer
focuses on and can be incomplete, biased, or skewed in its representation.
 ARTICLE IN PRESS

Uncertainty visualization errors 11

A visual description allows for enough understanding to mentally transform,

interpret, and make decisions with the representation. These processes require
the cognitive ability termed mental imagery, and again depend on working
memory (for a review, see Kosslyn, 1995).

2.4 Graph schemas

Graph schemas (third box in Fig. 3) are templates, rules, graphic conventions,
or strategies that people use to interpret a visualization. People might
develop graph schemas during formal education if they were taught how
to read different visualizations. For example, teachers commonly instruct
students on how to read maps by explaining the purpose and use of a legend,
scale, and compass rose. As another example, most educational institutions
introduce the number line to students at a young age, cementing a more is to the
right mapping (Winter & Matlock, 2013). These formal educational experi-
ences likely establish many graph schemas. People may also develop a graph
schema implicitly through practice viewing visualizations that utilize the same
conventions. Errors may occur with graph schemas if a viewer has not devel-
oped the necessary schema to interpret a visualization. A viewer may lack
graphic education or familiarity with a visualization, resulting in individual
differences in graph literacy (Okan, Garcia-Retamero, Galesic, & Cokely,
2012). Alternatively, the visualization might be wholly novel and require
the development of a new schema.

2.5 Matching process

The matching process between the visual description and the graph schema
occurs when a viewer selects a graph schema to expedite the process of inter-
preting a visualization. The mechanism of how viewers select a particular
schema remains unclear. Viewers may select a schema from the same broad
category as a visualization. For example, a viewer may select a schema for a
Cartesian coordinate plane when viewing a line graph. Viewers may also
select schemas based on matching salient features of the visualization and
the schema. When viewers see a bar chart, they may select a schema with
similar rectangular objects. Viewers may select the schema that is easiest
to recall, more recently stored, or one that has been primed. Errors may
occur when viewers select the wrong schema for a visualization. For exam-
ple, recent work shows that researchers could prime the type of schema par-
ticipants used by telling them an interesting story about the data (Xiong, Van
Weelden, & Franconeri, 2019). Different groups of people received
 ARTICLE IN PRESS

12 Lace Padilla et al.

different stories, and each story pointed out different features within the
same visualization. When asked what other people might see as essential
features in the data, participants were more likely to report that other people
would see the information they were primed to think of as relevant (Xiong
et al., 2019).

2.6 Instantiated graph schema

Instantiating the graph schema occurs when viewers update their mental rep-
resentation of the visualization to include information from the graph
schema. Errors can occur in this process if viewers select the wrong schema;
they can also occur if the viewers have to perform a complex mental transfor-
mation to update their mental representation with the information from the
schema. With a large mismatch between the schema and the viewer’s mental
representation of the visualization, a mental transformation may be required to
combine the two (Vessey, 1991). For example, if the Y-axis is inversely
ordered (low numbers at the top), the viewer may need to mentally transform
the visualization to correctly order the Y-axis (according to their schema)
before incorporating the visual description and schema. Increased errors
and time to instantiate the graph schema will occur in cases in which a large
mismatch and exorbitant mental transformations are required. Theses out-
comes result from overloading the limited resources of working memory
available as time decays the information and exceeds the capacity to hold
chunks of information in memory (Doumont, 2002).

2.7 Message assembly

The message assembly process describes how viewers interpret their mental rep-
resentation of the visualization after it has been updated by the graph schema.
The resultant conceptualization of the meaning of the graphic is the concep-
tual message (fourth box in Fig. 3). Errors that may occur at this stage of the
process result from taking the wrong meaning from the visualization’s men-
tal representation.

2.8 Conceptual question

The conceptual question (box below working memory in Fig. 3) refers to the
question that the viewer asks of the visualization. A viewer may have specific
goals, such as attempting a data analytics task, which could produce various
direct conceptual questions, as in “Where are the outliers?” or “Which vari-
ables have meaningful relationships?.” Conceptual questions can also be
 ARTICLE IN PRESS

Uncertainty visualization errors 13

more general (e.g., What can this visualization tell me about my health?)
or ill-defined (e.g., What am I looking at?). Many times, viewers may have
a sequence of conceptual questions about the visualization, which may
evolve.
In the Padilla, Creem-Regehr, Hegarty, & Stefanucci, 2018 framework,
conceptual questions play a key role as they channel working memory. This
framework suggests that the central executive (i.e., the resource allocation
mechanism in working memory) applies working memory to answer the
conceptual question during visualization reasoning. As a result, the concep-
tual question can:
1. Drive a viewer’s top-down attention to relevant information
2. Guide which graph schemas are selected
3. Frame the conceptual message
4. Influence decisions
The viewer’s specific question can influence all of the processes in this model
except bottom-up attention. These processes can also form feedback loops
or prime a specific graph schema (e.g., Xiong et al., 2019). Based on the con-
ceptual message, a viewer may decide to update the question or goal and
repeat some of the processes. Errors can occur as a result of the conceptual
question if it is unclear to the viewer how to achieve his or her particular
goals. The viewer might ask the wrong question to achieve their goals or
use incorrect steps. Viewers might also have too many goals, which can
be challenging to keep track of and require a significant amount of working
memory to manage.

2.9 Decision-making
Once all the relevant conceptual questions have been answered for the viewer
to feel comfortable making a decision, he or she completes the decision step.
The majority of the widely documented decision-making biases and heuristics
occurs in the decision-making step. This process involves taking the visual
information stored in the mind and using Type 1 or Type 2 processing to reach
a conclusion, usually in order to perform an action (Kahneman, 2011). Type 1
processing is relatively fast, unconscious, and intuitive. Type 2 processing
involves working memory and is slower, more metabolically intensive, and
more contemplative than Type 1 processing (Evans & Stanovich, 2013).
Other models of decision-making characterize these processes differently.
Here we note two processes in line with Evans and Stanovich (2013), one that
requires the activation of working memory to make a decision and another
 ARTICLE IN PRESS

14 Lace Padilla et al.

process that does not require significant working memory. There exists a mas-
sive body of literature detailing numerous possible decision-making biases that
can occur at this stage. Not all decision-making biases have been generalized
to the context of decisions with visualizations, but many of these biases may
influence reasoning with visualizations. However, more work is needed to
examine if all previously documented decision-making biases generalize to
the context of decision-making with visualizations.

2.10 Behavior
The final stage of the Padilla et al. (2018) model results in action or behavior.
Errors, although not decision-making errors, might occur in this model’s
final stage when people cannot take the action that they have selected.
For example, in hurricane forecasting, people might see a hurricane visual-
ization, decide to evacuate, and then lack the necessary resources to evacuate
or not know the appropriate evacuation route. These phenomena require
exploration in the more applied social sciences and are beyond the scope
of this chapter. However, failures to suppress heavily automated behaviors
(e.g., in the case of addictions) due to reduced cognitive resources or poor
executive control can also be observed during this stage.

3. Uncertainty visualization errors

Errors in understanding uncertainty visualizations can occur through-
out the decision-making process. Here we will use the Padilla et al. (2018)
cognitive model to organize and describe the widely documented errors as
early-, middle-, and late-stage visualization processing errors (as seen in
Fig. 5).

3.1 Early-stage processing errors

Early-stage visual processing errors are driven by the visual system and atten-
tional processes. Occurring early in the decision-making process, these types
of errors can be particularly hard to overcome as they influence all of the
downstream processes. Researchers refer to early-stage errors driven by ele-
ments in the visualization as visual-spatial biases (Padilla et al., 2018).
Researchers also speculate that visual-spatial biases are particularly hard to
overcome because they may be due to bottom-up attention and Gestalt
principles, both of which are difficult to cognitively control.
 ARTICLE IN PRESS
Fig. 5 Early-, middle-, and late-stage uncertainty visualization errors that are organized using the Padilla et al. (2018) visualization
decision-making model.
 ARTICLE IN PRESS

16 Lace Padilla et al.

3.1.1 Boundaries 5 conceptual categories

One visual-spatial bias that occurs often in uncertainty visualization arises when
visualization creators delineate continuous data with isocontours, boundaries,
intervals, bins, or other types of segmenting marks. Throughout our daily lives,
we have learned to interpret the delineations in the world, such as fences, road
lanes, and crosswalks, as indications of important information. Tversky (2011)
writes, “Framing a picture is a way of saying what is inside the picture has a
different status than outside of the picture” (p. 522). In our continuous world,
physical delineations separate and categorize meaningful differences and space
(Tversky, 2001). Delineations can also be metaphorical. We draw a proverbial
line in the sand to indicate a boundary that should not be crossed (Lakoff &
Johnson, 1980). As humans, we are adept at categorizing complex information,
and we commonly do this by physically or mentally constructing boundaries.
The problem for uncertainty visualization emerges when the designer
creates boundaries in probabilistic data, and the boundaries do not indicate
categorically different information (Padilla et al., 2015, 2017). For example,
95% confidence intervals delineate probabilistic information to indicate that
the true mean has a 95% chance of falling within the specified range.
However, there is no categorical difference between the data inside and out-
side of the confidence interval. Said another way, 95% confidence is not
unique, and scientists could have also chosen intervals at 96%, 94%, or
99%. Ninety-five percent confidence exists as a convention concerning
the probability of error scientists consider acceptable to make certain infer-
ences. Some fields have different conventions. The National Hurricane
Center uses a 66.66% confidence interval to communicate the uncertainty
in a hurricane forecast path.
When most viewers see an interval, they utilize the strategies they have
developed throughout their lives and interpret it as a meaningful boundary
that notes categorically different information (Tversky, Corter, Yu,
Mason, & Nickerson, 2012; Zacks & Tversky, 2013). This error could also
be considered a mismatch of the visual description and the instantiated graph
schema. In a geospatial context, researchers have called this a containment
strategy (McKenzie, Hegarty, Barrett, & Goodchild, 2016), where areas
within a boundary are imbued with semantic homogeneity (Fabrikant &
Skupin, 2005). For example, navigation applications show a user’s location,
but sometimes the location can have uncertainty (i.e., if the GPS signal is
interrupted). One study examined different visualizations for representing
the uncertainty in the viewer’s location by comparing a gradient map to a
95% CI (see Fig. 6). When viewing the 95% confidence interval that looked
 ARTICLE IN PRESS

Uncertainty visualization errors 17

Fig. 6 Visualizations that show the uncertainty in two locations, using a gradient or a
bounded circle right, used in McKenzie et al. (2016). Reproduced per CC-BY license, from
Padilla, L., Creem-Regehr, S., Hegarty, M., & Stefanucci, J. (2018). Decision making with
visualizations: A cognitive framework across disciplines. Cognitive Research: Principles
and Implications, 3, 29.

like a bounded circle, participants were more likely to take a containment

strategy than when viewing the same positional uncertainty represented
in a gradient (McKenzie et al., 2016).
Work examining hurricane forecasts also finds that people use a contain-
ment strategy. In one study, researchers showed participants five visualiza-
tions of a hurricane’s forecasted path (Fig. 7). The path visualizations were
intended to show the forecasted direction of the storm and the uncertainty in
the forecasted route. As the time increases from the initial forecast, it becomes
increasingly more difficult to accurately predict the path of the storm, which is
shown in the visualizations’ spread increasing for B–E in Fig. 7. Researchers
compared a version of the cone of uncertainty, which shows the main forecast
path of the storm, along with a 66.6% confidence interval (C in Fig. 7), to a
version with just the center line (A), a cone with no center line (B), a gradient
mapping of the confidence interval (D), and a new visualization technique
entitled an ensemble visualization (E). The ensemble visualization shows a subset
of paths sampled from the hurricane’s probabilistic forecast (Liu et al., 2016;
Liu, Padilla, Creem-Regehr, & House, 2019). This research demonstrated
that the visualizations that were cone-like (B–D in Fig. 7) elicited a contain-
ment strategy where participants rated areas inside of the cones to have more
damage than areas outside of the cones. With the ensemble visualization, par-
ticipants reported that areas near the center of the distribution would receive
more damage and damage ratings decreased along with the distance to the
center of the distribution (Ruginski et al., 2016). The response patterns
observed for ensemble visualizations indicate that participants understand
the distribution of uncertainty that the ensembles represent. In the context
of hurricane forecasting, this experiment was the first to find an alternative
 ARTICLE IN PRESS
Fig. 7 Redrawn versions of hurricane forecast path of visualizations based on Ruginski et al. (2016).
 ARTICLE IN PRESS

Uncertainty visualization errors 19

to the cone of uncertainty that did not elicit the containment strategy.
Researchers determined that the edge of the hard boundary elicits the highest
visual salience and likely drives the containment strategy (Padilla et al., 2017).
The misunderstandings associated with delineations occur in one-
dimensional data as well. Delineation errors can be understood as boundaries
creating conceptual categories (Padilla et al., 2021). The boundaries creating con-
ceptual categories error likely contributes to the numerous studies finding that
people misunderstand how to interpret error bars and confidence intervals.
Both well-trained experts in statistics and novices commonly misunderstand
how to interpret statistical significance from frequentist 95% confidence
intervals (e.g., Belia et al., 2005; Hofman, Goldstein, & Hullman, 2020).
Researchers find that even trained experts incorrectly assume that no signif-
icant difference exists between two groups with overlapping intervals (Belia
et al., 2005). When comparing two health treatments with visualized means
and frequentist 95% confidence intervals, participants were more willing to
overpay for treatment and to overestimate the effect size compared to when
the same data were shown with predictive intervals (Hofman et al., 2020).
People tend to believe that error bars contain the distribution of values,
resulting in the mismatch between the visual description and instantiated
graph schema. If the two bars are far apart, the boundaries lead people to
believe that these boundaries contain all the relevant values and therefore
they incorrectly assume a statistically significant difference. A similar effect
has also been found with bar charts. Researchers have demonstrated a
“within the bar bias,” where people believe that data points that fall within
a bar are more likely to be part of a distribution than data points equal dis-
tance from the mean but outside of the bar (Newman & Scholl, 2012).
This boundaries- create-conceptual-categories error likely occurs early in the
decision-making process. As demonstrated in Padilla et al. (2017), bound-
aries make up some of the most salient features in a visualization and can
attract our bottom-up attention. As a result, we might spend more time
looking at the boundaries in a visualization, which can produce an over-
weighting of the boundaries in our conceptualization of the data.
One of the reasons boundaries create conceptual categories is that they
may reinforce Gestalt grouping principles, which are the visual system’s pro-
pensity to group and categorize visual information based on similarities in
properties such as shape, color, physical proximity, and other contextual
information (Wertheimer, 1938). As an illustration, try to determine if pat-
terns are depicted in Fig. 8. All the figure items may seem to be a part of one
global grouping because they are all circular and loosely arranged in a circle.
 ARTICLE IN PRESS

20 Lace Padilla et al.

Fig. 8 Ambiguous Gestalt grouping example.

Fig. 9 Ambiguous Gestalt grouping example with boarder around the ovular items.

With effort, most viewers notice that some objects are larger or smaller and
others circular or ovular. Identifying patterns becomes easier when bound-
aries are added, as in Fig. 9, which bounds the ovular items with a line.
When the boundaries are included, visually grouping the ovular objects
and noticing they have an upward trend is much easier. The boundary works
to precategorize some of the information for the visual system. Said another
way, the boundaries offload cognition on the visualization by categorizing
the objects before the visual system does. The categorization created by the
boundaries occurs early in the decision-making process and reduces a visual
system processing step. However, a problem arises when a viewer needs to
group different information than what the boundary contains. When viewing
Fig. 9, try to mentally group the smaller objects. Most people can successfully
group the smaller objects and see their trend, but this process requires
 ARTICLE IN PRESS

Uncertainty visualization errors 21

significant effort. Mentally grouping the small objects requires suppressing or

ignoring the grouping formed by the boundary, requiring additional working
memory.
The prominence of the boundaries on the patterns we can see in the data
illustrates a visual-spatial bias, where the boundary can lead viewers to see
different patterns within the same visualization or data. When visualization
designers use boundaries, they define the types of patterns that viewers can
see in their data. Their viewers will have difficulty seeing any other patterns
within the data. The immutable effect of the boundary can be problematic
when the boundaries are arbitrary (i.e., 95% CIs or 66% CIs in the Cone of
Uncertainty), making viewers believe that categorical differences exist in
the data when there are none, which hinders viewers from finding other
important patterns.
Although the point at which the boundary enters the decision-making
process occurs very early (e.g., the visual array), the impact of boundaries
might be observed at multiple points throughout the decision-making pro-
cess. Boundaries may be highly salient and direct viewers’ bottom-up atten-
tion to information inside the boundaries. Viewers might form a strategy to
assume that visualization designers are trying to communicate something
meaningful with the boundaries and direct their top-down attention to the
boundaries’ information. Boundaries may evoke incorrect schemas and lead
to misunderstandings about what the data represent. Boundaries could even
evoke some traditional decision-making biases such as anchoring, where peo-
ple are biased to make judgments in relationship to the boundaries.
Such early-stage processing errors are some of the most consistent and
widely documented, but little is known about why these errors occur.
One theory that we propose here is that working memory is a crucial con-
tributor to early-stage processing errors in uncertainty visualization.
Early-stage processing errors represent a unique category because working
memory cannot easily influence all these processes. In particular, bottom-up
attention is difficult to control with effort. As noted throughout this section,
many of the errors we reviewed might be fully explained by bottom-up
attentional processes. For example, some work finds that boundaries in hur-
ricane forecasts are highly salient and draw the viewers’ attention (Padilla
et al., 2017). All types of boundaries may draw viewers’ attention, and there-
fore, they have an overstated impact on viewers’ decision-making process
compared to more task-relevant information in the visualization. Further,
even when viewers consciously know not to focus on information, as with
the Cone of Uncertainty boundary, they likely have difficulty suppressing
 ARTICLE IN PRESS

22 Lace Padilla et al.

saccadic movements toward such salient information. Thus, the boundary

may increase the working memory required due to active inhibition.
The most notable characteristic of early-stage processing errors is that
they are challenging to overcome. For example, in one study, participants
were provided with extensive instructions on interpreting the Cone of
Uncertainty (Boone, Gunalp, & Hegarty, 2018). Researchers instructed
participants that the cone does not show the storm’s size growing over time,
but participants still made decisions as if the storm’s size was increasing.
Notably, at the end of the experiment, participants could accurately answer
questions about interpreting the cone correctly (Boone et al., 2018). This
work provides some evidence that even when viewers are aware that they
should cognitively override the visual array’s impacts, they find it challeng-
ing to do so.
Participants’ inability to utilize working memory to make more effective
decisions in the previous examples may be because working memory has
difficulty impacting early processing errors. Working memory’s problem
in affecting early processing errors could be due to earlier errors biasing
all the downstream processes. It could also be the case that early processing
errors are primarily due to bottom-up attention and working memory may
have little ability to impact bottom-up attention. However, no work has
examined the exact nature of early-stage processing errors in visualization
reasoning. More work is needed to understand the cause, prevalence, and
unwavering nature of such errors.

3.2 Middle-stage processing errors

In the Padilla et al. (2018) framework, middle-stage processing errors occur
after the visual system has created a mental representation of the visualization.
At this stage, viewers apply a schema that they have stored in long-term mem-
ory to their mental representation of the visualization. For example, when
viewing Fig. 10 (left), most people would categorize the picture as a map.
Consciously or unconsciously, they would retrieve the schema for maps
and make assumptions about the information, including that North is at
the top and that a consistent relationship likely exists between the physical size
of the areas shown. They would have made assumptions based on map
schemas even though we excluded the map’s compass rose and legend. As
in this example, many of the assumptions we make about visualizations based
on schemas create an advantage over their absence. Schemas help us interpret
information correctly, efficiently, and quickly when a visualization adheres to
known graphic conventions.
 King’s Cross King’s Cross
St Pancras St Pancras

Edgware Euston
Paddington Road Baker St.
Euston

ARTICLE IN PRESS
Baker St. Old Street
Warren
Warren Marylebone St. Farringdon
Moorgate
St. Liverpool
Street
Liverpool St.
Edgware Tottenham Holborn
Road Oxford Court Road
Circus Moorgate
Bank Notting Bond Oxford
Aldgate
Paddington Hill Gate Street Circus Holborn
Leicester
Bond St. Square Tower Hill
Notting Bank
Hill Gate Embankment London
Piccadilly Bridge Aldgate
Green Park Circus Charing Green Park Cannon
Waterloo
Cross Leicester Street
Square
Piccadilly
Circus Monument Tower Hill
High St. Charing
Kensington Cross
Westminster
Gloucester Elephant
Gloucester Blackfriars
Road & Castle
Road
Earl’s
Court
South Victoria Westminster Embankment London Bridge
Kensington Earl’s South Victoria
Court Kensington

Waterloo

Geographically Accurate London Tube Map Diagrammatic London Tube Map

Fig. 10 Left geographically accurate transit map and right diagrammatic map of the London Tube. Redrawn from Guo, Z. (2011). Mind the map!
The impact of transit maps on path choice in public transit. Transportation Research Part A: Policy and Practice, 45(7), 625–639.
 ARTICLE IN PRESS

24 Lace Padilla et al.

The Padilla et al. (2018) schema instantiation process has three steps, as
illustrated in Fig. 11. A viewer must first correctly classify the visualization
type. With standard visualizations (e.g., line or bar charts), accurate classifi-
cation occurs relatively easily. However, errors can arise when ambiguity
exists in classifying the category or type for a visualization.
A famous example of classification involves the London Underground map
by Harry Beck (see an example based on Beck’s innovation in Fig. 10, right).
Beck helped define a new cartographic convention that departed from the his-
torical approach of superimposing subway lines on a geographically accurate
map (Guo, 2011). In Beck’s redesign, he opted to arrange the layout in a dia-
grammatic fashion that focused on improving the legibility of routes, transfers,
and stops, inspired by electrical circuits. Initially, transit officials scoffed at the
design, but it was ultimately adopted in 1933. Some of the apprehension about
Beck’s map began because officials thought that riders might see it as a standard
map, fail to realize that the distances between stops were not based on physical
distance, become confused, and miss their stops. Researchers continue to
discuss whether Beck’s design should be classified as a map or as a diagram
(Cartwright, 2012).
When new innovations change visualization design, viewers might
become confused about how to classify a new type of visualization, which
can affect how they determine and implement an appropriate schema.
Today, Beck’s approach has been utilized worldwide for close to a century,
and most transit riders have developed a specific schema for diagrammatic
subway maps. Beck’s success is likely due in part to the design being different
enough from standard approaches that the design prompted riders to recog-
nize that a standard map-based schema would not work. Additionally, the
design reduced directional information to three axes, reducing the memory
required to match viewers’ destination goals with their visual description.
In the next step of the graph instantiation process, viewers retrieve the
relevant schema based on how they classified the visualization. Errors can
occur in this process when viewers have not learned an appropriate schema.
When no schema is available for a graph type, the viewers might utilize a
schema from a different visualization type or context. For example, see
the new coordinate system in Fig. 12 and try to determine the values for B.

Visual description Schema retrieval

(from long-term Schema application
classification
knowledge)

Fig. 11 Three-step schema instantiation process.

 ARTICLE IN PRESS

Uncertainty visualization errors 25

A
(2,3)

B
(?,?)

Fig. 12 Hypothetical new coordinate plane.

Fig. 13 Example of a mental rotation needed to apply at the schema for Cartesian coor-
dinate plane to a hypothetical new coordinate plain and derived B.

One strategy is to notice that A and B both have two values and a coor-
dinate plane. Dot plots use similar Cartesian coordinate planes but have dif-
ferent axes than in the example. One could apply the Cartesian coordinate
schema to interpret the new hypothetical coordinate plane and then derive
B’s values, as illustrated in Fig. 13.
The problem with applying the schema for a Cartesian coordinate plane to
the new coordinate plane is that the planes do not adhere to the same graphic
conventions. The angles of the axes in Fig. 14 are not 90°. Applying a schema
for a Cartesian coordinate to the new coordinate plane incorrectly is easy,
as they share similar properties. When the appropriate schema is unknown,
viewers commonly retrieve a different visualization schema to interpret the
new information, which can work out well in some cases or can lead them
to systematic misinterpretations. Graph schemas that viewers can easily
remember and those frequently used are more likely to be applied to an
ambiguous visualization type.
In the final stage of the schema instantiation process, viewers must apply
the schema that they have retrieved to the visualization in order to answer
 ARTICLE IN PRESS

26 Lace Padilla et al.

A
(2,3)

B
(-2,3)

Fig. 14 Illustration of how the axes for the new coordinate plane are not 90° angles.

the conceptual question. When a mismatch between the schema and the
visualization occurs, as illustrated in the prior example, a transformation is
required to make the two align. Cognitive Fit Theory describes how errors
occur when a mismatch between the schema and the visualization requires
exorbitant mental computations (Vessey, 1991). A large mismatch between
the schema and visualization requires significant working memory to make
the two align, which results in increased errors and time to complete the task
(Padilla et al., 2018). Note that the Padilla et al. (2018) model suggests that
the schema matching process and all other processes (other than bottom-up
attention) are in service of the conceptual question. Even if viewers do not
think they are trying to answer a specific question, they always have a goal,
which could be as simple as understanding what they see.

3.2.1 Schema errors in hurricane visualizations

Uncertainty visualizations of hurricane forecasts represent one of the most
highly studied types of schema errors (Padilla et al., 2017; Padilla, Creem-
Regehr, et al., 2020; Ruginski et al., 2016). As previously discussed, viewers
assume that the National Hurricane Center’s Cone of Uncertainty represents
the storm’s size growing over time, even though it does not communicate
storm size information (Padilla et al., 2017). Researchers have also observed
the misunderstanding that the cone’s area represents the size of the storm
when blurry or fuzzy boundaries border the cone.
One key source of these errors involves the schema that people utilize
when seeing hurricane forecast maps. Viewers looking at a hurricane forecast
map reasonably use the schema that they have learned for maps, which dic-
tates that physical distance on a map should correspond to physical distance
 ARTICLE IN PRESS

Uncertainty visualization errors 27

in the world. However, cone-like hurricane forecasts violate cartographic

expectations by using physical distance to represent uncertainty in the
storm’s path. To interpret the forecast correctly, viewers must maintain
the base-map schema but then suppress the map schema when looking at
the cone. Viewers must keep schemas for both maps and uncertainty in
working memory and apply each where appropriate in the visualization.
Flexibly switching between schemas is highly demanding on working mem-
ory. Such a high working memory demand required by cone-like hurricane
forecasts may overtax viewers’ limited working memory capacity.
When the working memory demand of a visualization exceeds a viewer’s
working memory capacity, the viewer may drop one schema (e.g., use only
the map schema). Viewers who utilize only a map schema commonly report
that the cone-like visualizations represent a danger zone, where areas inside
the cone are at risk and areas outside the cone are relatively safe (Ruginski
et al., 2016). When forced to drop a schema, we argue that people will likely
maintain the schema with which they have the strongest associations. As
most people have seen and used maps for large portions of their lives, the
map schema will take prominence over the uncertainty schema, which they
may have less training or experience using.
We were initially surprised to find that viewers of a blurry cone also see a
similar danger zone, as researchers have suggested that blur/fuzziness/trans-
parency may be a more intuitive way to communicate uncertainty
(MacEachren et al., 2012). Researchers continue to test blurry cones as an
alternative approach to the Cone of Uncertainty and see no benefits of blur
(Millet et al., 2020). The interest in testing alternative metaphorical expres-
sions of uncertainty (e.g., blur, fuzziness, transparency, fogginess, and sketch-
iness), including our own, occurred mainly due to a misattribution of why
the Cone of Uncertainty leads to misunderstandings. We argue that the prin-
cipal error inherent in cone-like visualizations is that they force viewers to
hold multiple schemas in working memory, which is the case for cones with
both rigid and blurry boundaries. Blur, fuzziness, transparency, fogginess, and
sketchiness express uncertainty explicitly as an additional attribute of the visu-
alization that requires a second schema. More modern uncertainty visualiza-
tion techniques implicitly communicate the uncertainty in animations
(Hullman, Resnick, & Adar, 2015) or color (Correll, Moritz, & Heer,
2018) and may prove to be more effective because they do not require the
viewer to hold multiple schemas in their mind.
Blur or distributional visualizations can be highly successful if a second
schema is not required to understand the visualization. For example,
 ARTICLE IN PRESS

28 Lace Padilla et al.

researchers have found that gradient plots of 1D data can outperform interval
plots of the same information (Correll & Gleicher, 2014). Gradient plots of
1D data require only a single schema for mapping opacity to probability.
Researchers have also documented how ensemble visualizations, which
are the most effective hurricane forecast visualization technique (Ruginski
et al., 2016), can also suffer from schema errors (Padilla et al., 2017;
Padilla, Creem-Regehr, et al., 2020). The approach of this work was to
identify the schema participants use when viewing ensemble visualizations.
Ensemble visualizations have been developed as a technique relatively
recently (Liu et al., 2016), and we can reasonably assume that people have
not developed a specific schema for ensembles.
After reviewing all commonly available visualization techniques,
researchers noted that the ensemble visualization shared many similar prop-
erties to map-based navigation applications (Padilla et al., 2017). Both
map-based travel applications and ensemble hurricane forecasts have a base
map that adheres to standard cartographic principles and overlays of lines.
Researchers speculated that when viewing an ensemble visualization, people
utilize the schema that they have developed for understanding travel appli-
cations (Padilla et al., 2017). An essential benefit to using a travel application
schema is that participants would not have to hold multiple schemas in their
minds (e.g., one for maps and one for uncertainty). The use of a single
schema could be one reason why ensemble visualizations outperform
cone-like hurricane forecasts (Padilla et al., 2017; Ruginski et al., 2016).
However, the problem with using a travel application schema for hur-
ricane ensembles is that the schema could lead to errors in specific cases.
Researchers tested an additional hypothesis that people see each line of
the hurricane forecast ensemble as a specific path the hurricane could take
(Padilla et al., 2017). The schema for geospatial travel visualizations dictates
that the application shows a finite list of possible discrete routes and not a
distribution of routes. Whereas for the ensemble visualization, each line
depicts a subset of a distribution. In other words, the ensemble lines show
the spread of uncertainty in the path of the storm. They do not show an
exhaustive list of every possible path the storm could take. If people use a
schema for geospatial travel applications and one of the ensemble members
intersects a location of interest, they may incorrectly think the likelihood is
higher that the storm will hit that location (Padilla et al., 2017).a
a
Note that researchers provided participants little information about how to interpret the ensembles,
which simulates the conditions in which they would see hurricane forecast in the news (i.e., on average,
hurricane forecasts are shown on TV for 1.52 min; Padilla, Creem-Regehr, et al., 2020; Padilla,
Powell, Kay, & Hullman, 2020).
 ARTICLE IN PRESS

Uncertainty visualization errors 29

Researchers tested the hypothesis that viewers use a geospatial travel

application schema to interpret ensemble displays by showing participants
ensemble hurricane forecasts with two indicated locations (See Fig. 15)
(Padilla et al., 2017). On each trial, an ensemble member intersected one
of the locations. Researchers found that when the ensemble member inter-
sected a location, the participants believed that the location would receive
more damage than the location that was not intersected by an ensemble
member. This overreaction due to the colocation with the ensemble mem-
ber persisted regardless of the damage probability (Padilla et al., 2017).
Follow-up research provided converging evidence that ensemble visu-
alizations evoke a geospatial travel application schema by replicating the
overreaction when an ensemble member intersects a point of interest and
demonstrating that the number of lines shown moderates this effect
(Padilla, Creem-Regehr, et al., 2020). Researchers reduced the overreaction
bias by increasing the number of lines shown. As an illustration, when shown
an ensemble visualization with 5 or 10 paths and one path intersects a loca-
tion, people commonly report a 20% chance the storm will hit the location
with 5 paths and 10% with 10 paths. Researchers found that increasing the
lines from 9 to 14 to 33 meaningfully reduced the overreaction bias (Padilla,
Creem-Regehr, et al., 2020).
However, researchers were never able to entirely eliminate the
overreaction by changing the number of lines (Padilla, Creem-Regehr,
et al., 2020). In a final attempt to reduce the overreaction bias, researchers

Fig. 15 Example ensemble hurricane forecast visualizations with two locations from
Padilla et al. (2017). In each visualization, one location is intercepted by an
ensemble member. Reproduced per CC-BY license, from Padilla, L., Creem-Regehr, S.,
Hegarty, M., & Stefanucci, J. (2018). Decision making with visualizations: A cognitive frame-
work across disciplines. Cognitive Research: Principles and Implications, 3, 29.
 ARTICLE IN PRESS

30 Lace Padilla et al.

tested if participants could override the graph schema using working mem-
ory to control their overreaction cognitively. Researchers provided partic-
ipants with extensive instructions on interpreting an ensemble visualization
and how to perform the task correctly. Participants with extensive instruc-
tions were able to reduce their bias but not entirely remove it. At the end of
the study, participants who received extensive instructions could report the
correct strategy, but these participants still overreacted in their behavioral
judgments, albeit to a lesser degree (Padilla, Creem-Regehr, et al., 2020).
In summary, ongoing research on hurricane forecast visualizations dem-
onstrates multiple schema-related errors. Errors are highly likely when
working memory demand from a visualization is increased, by maintaining
two schemas or attempting to cognitively override one schema. The major-
ity of geospatial uncertainty visualizations will likely encounter similar errors
because superimposing the uncertainty visualization on the base map will
likely evoke the viewer’s map schema.
Future visualization designers interested in communicating geospatial
uncertainty that does not evoke a traditional cartographic schema could uti-
lize the approach pioneered by Harry Beck in the London Underground
map. One possible reason that the London Underground map does not pro-
duce large schema-based errors is that its differences sufficiently separate the
visualization from a traditional map, which makes people aware that a con-
ventional map schema is not appropriate. If the visualization alerts the viewer
to its novelty, it could trigger the viewer to develop a new schema.

3.2.2 Deterministic construal errors

Many schema-based errors may also be explained by viewers ignoring the
uncertainty and instead interpreting uncertainty visualizations as communicat-
ing deterministic data, called deterministic construal errors ( Joslyn & Savelli,
2020). Researchers first identified deterministic construal errors in 1D tem-
perature forecasts, when they presented participants with uncertainty in mean
temperature forecasts with confidence intervals visualized as bars (Savelli &
Joslyn, 2013). The researchers found that 36% of participants believed that
the confidence intervals represented high- and low-temperature forecasts
rather than uncertainty around the mean (Savelli & Joslyn, 2013). Savelli
and Joslyn then tested alternative visualization techniques, including dotted
lines and blurry boundaries, and found that the participants still assumed that
the intervals around the means were high- and low-temperature forecasts.
The researchers went further by creating an obvious key that instructed
Another Random Scribd Document
with Unrelated Content
of years could not abate the anger of a man like him. Mrs. Hemson
was dead now, a twelvemonth past; so that I was entirely alone in
the world. As to the will, it had not been found, as was to be
supposed, or the money would have been mine. My growth in years,
the passing from the little girl into the woman, and the new ties and
interests of my foreign school life had in a degree obliterated those
unhappy events, and I scarcely ever gave even a thought to the
past.
Mr. and Mrs. Paler were staying temporarily at Nulle; well-
connected English people, about to fix their residence in Paris. They
were strangers to me personally, but the Miss Barlieus knew
something of their family, and we heard that Mrs. Paler was inquiring
for a governess; one who spoke thoroughly English, French, and
German. Mademoiselle Annette thought it might suit me, and
proposed to take me to call on them at the Lion d'Or hotel.
I seized upon the idea eagerly. The word Paris had wrought its
own charm. To be conveyed to that city of delight appeared only
secondary to entering within the precincts of a modern Elysium.
"Oh, Mademoiselle Annette, pray let us go! I might perhaps do for
them."
Mademoiselle Annette laughed at the eagerness so unequivocally
betrayed. But she set off with me the same day.
The Lion d'Or was full. Mr. and Mrs. Paler had no private sitting-
room (there were only two salons in the whole house), and we were
ushered into their chamber, French fashion. Mr. Paler was a stout
man in gold spectacles, shy and silent; his wife, a tall handsome
woman with large eyes and dark hair, talked enough for both. Some
conversation ensued, chiefly taken up by Mrs. Paler explaining the
sort of governess she wished for, Mr. Paler having quitted us.
"If you require a completely well-educated young lady—a
gentlewoman in every sense of the term—you cannot do better than
engage Miss Hereford," said Mademoiselle Annette.
"But what's her religion?" abruptly asked Mrs. Paler. "I would not
admit a Roman Catholic into the bosom of my family; no, not though
she paid me to come. Designing Jesuits, as a great many of them
are!"
Which, considering she was speaking to a Roman Catholic, and
that a moment's consideration might have told her she was, evinced
anything but courtesy on the lady's part, to say nothing of good
feeling. Mademoiselle Annette's brown cheek deepened, and so did
mine.
"I belong to the Church of England, madam," I answered.
"And with regard to singing?" resumed Mrs. Paler, passing to
another qualification unceremoniously. "Have you a fine voice?—a
good style?—can you teach it well?"
"I sing but little, and should not like to teach it. Neither am I a
very brilliant player. I have no great forte for music. What I do play I
play well, and I can teach it well."
"There it is! Was there ever anything so tiresome?" grumbled Mrs.
Paler. "I declare you cannot have everything, try as you will. Our last
governess was first-rate in music—quite a divine voice she had—and
her style perfect; but, of all the barbarous accents in French and
German (not to speak of her wretched grammar), hers were the
worst. Now, you are a good linguist, but no hand at music! What a
worry it is!"
"May I ask what age your children are?" interposed Mademoiselle
Annette, who could speak sufficient English to understand and join
in the conversation.
"The eldest is twelve."
"Then I can assure you Miss Hereford is quite sufficient musician
for what you will want at present, madam. It is not always the most
brilliant players who are the best instructors; our experience has
taught us the contrary is the case."
Mrs. Paler mused. "Does Miss Hereford draw?"
"Excellently well," replied Mademoiselle Annette.
"I have a great mind to try her," debated Mrs. Paler, as if
soliloquizing with herself. "But I must just pay my husband the
compliment of asking what he thinks: though I never allow any
opinion of his to influence me. He is the shyest man! he went out,
you saw, as you came in. I am not sure but he will think Miss
Hereford too good looking; but she has a very dignified air with her,
though her manners are charmingly simple."
"When you have considered the matter, madam, we shall be glad
to receive your answer," observed Mademoiselle Annette, as she
rose. And Mrs. Paler acquiesced.
"Anne," began Mademoiselle Annette, as we walked home, "I do
not think that situation will suit you. You will not be comfortable in
it."
"But why?" I asked, feeling my golden visions of Paris dimmed by
the words. "I think it would perfectly suit me, Mademoiselle."
"Madame Paler is not a nice lady; she is not a gentlewoman. I
question, too, if she would make you comfortable."
"I am willing to risk it. You and Mademoiselle Barlieu have told me
all along that I cannot expect everything."
"That is true, my child. Go where you will, you must look out for
disagreeables and crosses. The lives of all of us are made up of
trials; none, save ourselves, can feel them; few, save ourselves, can
see, or will believe in them. Many a governess, tossed and turned
about in the world's tempest, weary of her daily task, sick of its
monotony, is tempted, no doubt, to say, 'Oh that I were established
as the Demoiselles Barlieu are, with a home and school of my own!'
But I can tell you, Anne, that often and often I and my sister envy
the lot of the poorest governess out on her own account, because
she is free from anxiety."
She spoke truly. Every individual lot has its peculiar trials, and
none can mitigate them. "The heart knoweth its own bitterness." I
walked on by her side then, in my young inexperience, wondering
whether all people had these trials, whether they would come to me.
It was my morning of life, when the unseen future looks as a bright
and flowery dream. Mademoiselle Annette broke the silence.
 "You will never forget, my dear, that you have a friend in us.
Should you meet with any trouble, should you be at any time out of
a situation, come to us; our house is open to you."
"Thank you, thank you, dear Mademoiselle Annette," I replied,
grasping her hand. "I will try and do brave battle with the world's
cares; I have not forgotten my mother's lessons."
"Anne," she gravely responded, "do not battle: rather welcome
them."
Well, I was engaged. And, as the Demoiselles Barlieu observed, it
was not altogether like my entering the house of people entirely
strange, for they were acquainted with the family of Mr. Paler:
himself they had never before seen, but two of his sisters had been
educated in their establishment.
A week or two after the Palers had settled themselves in Paris, I
was escorted thither by a friend of the Miss Barlieus. The address
given me was Avenue de St. Cloud, Commune de Passy. We found it
a good-looking, commodious house, and my travelling protector,
Madame Bernadotte, left me at the door. A young girl came forward
as I was shown into a room.
"Are you Miss Hereford, the new governess?"
"Yes. I think I have had the pleasure of seeing you at Nulle," I
answered, holding out my hand to her.
"That I'm sure you've not. I never was at Nulle. It was Kate and
Harriet who went there with papa and mamma. I and Fanny and
Grace came straight here last week from England, with nurse."
Now, strange to say, it had never occurred to me or to the Miss
Barlieus to ask Mrs. Paler, during the negotiations, how many pupils
I should have. Two children were with them at Nulle, Kate and
Harriet, and I never supposed that there were others; I believed
these would be my only pupils.
"How many are you, my dear?"
"Oh, we are five.
"Am I to teach you all?"
 "Of course. There's nobody else to teach us. And we have two
little brothers, but they are quite in the nursery."
Had Mrs. Paler purposely concealed the number? or had it been
the result of inadvertence? The thought that came over me was,
that were I engaging a governess for five pupils, I should take care
to mention that there were five. They came flocking round me now,
every one of them, high-spirited, romping girls, impatient of control,
their ages varying from six to twelve.
"Mamma and papa are out, but I don't suppose they'll be long. Do
you want to see mamma?"
"I shall be glad to see her.
"Do you wish for anything to eat?" inquired Miss Paler. "You can
have what you like: dinner or tea; you have only to ring and order it.
We have dined and had tea also. Mamma has not; but you don't
take your meals with her."
As she spoke, some noise was heard in the house, and they all ran
out. It proved to be Mrs. Paler. She went up to her own sitting-room,
and thither I was summoned.
"So you have got here safely, Miss Hereford?" was her salutation,
spoken cordially enough. But she did not offer to shake hands with
me.
"I have been making acquaintance with my pupils, madam. I did
not know there were so many."
"Did you not? Oh, you forget; I have no doubt I mentioned it."
"I think not. I believed that the two Miss Palers I saw at Nulle
were your only children."
"My only children! Good gracious, Miss Hereford, what an idea!
Why, I have seven! and have lost two, which made nine, and shall
have more yet, for all I know. You will take the five girls; five are as
easily taught as two."
I did not dispute the words. I had come, intending and hoping to
do my duty to the very utmost extent, whether it might be much or
little. Though certainly the five pupils did look formidable in
prospective, considering that I should have to teach them
everything, singing excepted.
"I hope you will suit me," went on Mrs. Paler. "I have had many
qualms of doubt since I engaged you. But I can't beat them into Mr.
Paler; he turns round, and politely tells me they are 'rubbish,' as any
heathen might."
"Qualms of doubt as to my being but nineteen, or to my skill in
music?" I asked.
"Neither; your age I never made an objection, and I daresay your
music will do very well for the present. Here's Mr. Paler."
He came in, the same apparently shy, silent, portly man as at
Nulle, in his gold spectacles. But he came up kindly to me, and
shook hands.
"My doubts turn upon serious points, Miss Hereford," pursued Mrs.
Paler. "If I thought you would undermine the faith of my children
and imbue them with Roman Catholic doctrines——"
"Mrs. Paler!" I interrupted in surprise. "I told you I was a
Protestant, brought up strictly in the tenets of the Church of
England. Your children are of the same faith: there is little fear, then,
that I should seek to undermine it. I know of none better in the
world."
"You must excuse my anxiety, Miss Hereford. Can you
conscientiously assure me that you hate all Roman Catholics?"
I looked at her in amazement. And she looked at me, waiting for
my answer. A smile, unless I mistook, crossed the lips of Mr. Paler.
"Oh, Mrs. Paler, what would my own religion be worth if I could
hate? Believe me there are excellent Christians amidst the Roman
Catholics, as there are amidst us. People who are striving to do their
duty in this world, living and working on for the next. Look at the
Miss Barlieus! I love them dearly; every one respects them: but I
would not change my religion for theirs."
"Is it the fact of your having spent four years in their house that
makes me doubtful. But I think can trust you; you look so sincere
and true. The alarming number of converts to Romanism which we
have of late years been obliged to witness, must make us all
fearful."
"Perverts, if you please," interrupted Mr. Paler. "When I hear of our
folks going over to the Romish faith, I always suspect they are those
who have not done their duty in their own. A man may find all he
wants in his own religion, if he only looks out for it."
"Oh, that's very true," I exclaimed, my eyes sparkling, glad,
somehow, to hear him say it. "It is what I have been trying to
express to Mrs. Paler."
"She has got her head full of some nonsensical fear that her
children should be turned into Roman Catholics—I suppose because
we are in a Catholic country," he resumed, looking at his wife
through his glasses. "She'll talk about it till she turns into one
herself, if she doesn't mind; that's the way the mania begins. There's
no more fear of sensible people turning Catholics than there is of my
turning Dutchman: as to the children, the notion is simply absurd.
And what sort of weather have you had at Nulle, Miss Hereford,
since we left it?"
"Not very fine. Yesterday it poured with rain all day."
"Ah. That would make it pleasant for travelling, though."
"Yes: it laid the dust."
"Did you travel alone?"
"Oh, no; the Miss Barlieus would not have allowed it. It is not
etiquette in France for a young lady to go out even for a walk alone.
An acquaintance of Miss Barlieus, Madame Bernadotte, who was
journeying to Paris, accompanied me."
"Well, I hope you will be comfortable here," he concluded.
"Thank you; I hope so."
"And look here, I'll give you a hint. Just you get the upper hand of
those children at once, or you'll never do it. They are like so many
untrained colts."
 Nothing more was said. I had not been asked to sit, and supposed
the silence was a hint that I must quit the room. Before I had got
far, a servant came and said I was to go back to it. Mrs. Paler was
alone then, looking very solemn and dark.
"Miss Hereford, you have been reared in seclusion, mostly in
school, and probably know little of the convenances—the exactions
of social life. Do not be offended if I set you right upon a point—I
have no doubt you have erred, not from want of respect, but from
lack of knowledge."
What had I done? of course I said I should be obliged to her to
set me right in anything when found wrong.
"You are a governess; you hold a dependent situation in my
house. Is it not so?"
"Certainly it is," I answered, wondering much.
"Then never forget that a certain amount of respect in manner is
due to myself and to Mr. Paler. I do not, of course, wish to exact the
deference a servant would give—you must understand that; but
there's a medium: a medium, Miss Hereford. To you, I and Mr. Paler
are 'madam' and 'sir,' and I beg that we may be always addressed as
such."
I curtsied and turned away, the burning colour dyeing my face. It
was my first lesson in dependence. But Mrs. Paler was right; and I
felt vexed to have forgotten that I was only a governess. Misplaced
rebellion rose in my heart, whispering that I was a lady born; that
my family was far higher in the social world than Mr. or Mrs. Paler's;
whispering, moreover, that that lady was not a gentlewoman, and
never could be one. But after a few minutes spent in sober
reflection, common sense chased away my foolish thoughts, leaving
in place a firm resolution never so to transgress again. From that
hour, I took up my position bravely—the yielding, dependent,
submissive governess.
But what a life of toil I entered upon! and—where were my
dreams of Paris? Have you forgotten that they had visited me, in all
their beautiful delusion? I had not. Delusive hopes are always the
sweetest.
When I had stayed three months at Mrs. Paler's I had never once
been into Paris further than the Champs Elysées. Save that we went
every Sunday morning in a closed carriage to the Ambassador's
chapel, I saw nothing of Paris. The streets may have been of crystal,
the fountains of malachite marble, the houses of burnished gold, for
all I witnessed of them—and I believe my warm imagination had
pictured something of the like resplendence. There was no pleasure
for me; no going out; my days were one lasting scene of toil.
I am not going to complain unjustly of Mrs. Paler's situation, or
make it out worse than it was. It has become much the fashion of
late years—I may say a mania—to set forth the sorrows and ill-
treatment that governesses have to endure: were the other side of
the question to be taken up, it might be seen that ladies have as
much to bear from governesses. There are good places and there
are bad ones; and there are admirable governesses, as well as
undesirable and most incapable ones: perhaps the good and bad, on
both sides are about balanced. I was well-treated at Mr. Paler's; I
had a generous diet, and a maid to wait upon me in conjunction
with the two elder girls. When they had visitors in an evening, I was
admitted on an equality (at any rate to appearance); I had respect
paid me by the servants; and I was not found fault with by Mr. and
Mrs. Paler. Could I desire better than this? No. But I was
overworked.
Put it to yourselves what it was, if you have any experience in
teaching. Five girls, all in different stages of advancement, to learn
everything, from German and good English down to needle-work.
The worst task was the music; the drawing lessons I could give
conjointly. All five learnt it, piano and harp, and two of them, the
second and the youngest but one, were so wild and unsteady that
they could not be trusted to practise one instant alone. I rose every
morning at half-past six to begin the music lessons, and I was
usually up until twelve or one o'clock the next morning correcting
exercises, for I could not find time to do them during the day. "Make
time," says somebody. I could only have made it by neglecting the
children.
"Our last governess never did a thing after six in the evening,"
Kate said to me one day. "You should not be so particular, Miss
Hereford."
"But she did not get you on to your mamma's satisfaction."
"No, indeed: mamma sent her away because of that. She did not
care whether we advanced or not. All she cared for was to get the
studies over anyhow."
Just so: it had been eye-service, as I could have told by their
ignorance when I took the girls in hand. My dear mother had
enjoined me differently: "Whatever you undertake, Anne, let it be
done to the very best of your ability: do it as to God; as though His
eye and ear were ever present with you."
I appealed to Mrs. Paler: telling her I could not continue to work
as I was doing, and asking what could be done.
"Oh, nonsense, Miss Hereford, you must be a bad economizer of
time," she answered. "The other governesses I have had did not
complain of being overworked."
"But, madam, did they do their duty?"
"Middling for that—but then they were incorrigibly lazy. We are
quite satisfied with you, Miss Hereford, and you must manage your
time so as to afford yourself more leisure."
I suggested to Mrs. Paler that she should get help for part of the
music lessons, but she would not hear of it; so I had to go on doing
my best; but to do that best overtaxed my strength sadly. Mrs. Paler
might have had more consideration: she saw that I rarely went out;
one hurried walk in the week, perhaps, and the drive to church on
Sunday. My pupils walked out every day, taken by one or other of
the servants; but they did not go together: two or three stayed with
me while the rest went, and when they came back to me these
went. Mrs. Paler insisted upon my giving an hour of music to each
child daily, which made five hours a day for music alone. The
confinement and the hard work, perhaps the broken spirits, began to
tell upon me; nervous headaches came on, and I wrote to the Miss
Barlieus, asking what I should do. I wrote the letter on a Sunday, I
am sorry to say, failing time on a week day. None of us went abroad
on a Sunday afternoon. Mrs. Paler protested that nothing but sin and
gallivanting was to be seen out of doors on a French Sunday; and
once home from church we were shut up for the rest of the day. She
did not go out herself, or suffer anybody else to go; Mr. Paler
excepted. He took the reins into his own hands.
The Miss Barlieus answered me sensibly; it was Miss Annette who
wrote. "Put up with it to the close of your year from the time of
entrance," she said. "It is never well for a governess to leave her
situation before the year is up, if it can be avoided; and were you to
do so, some ladies might urge it as an objection to making another
engagement with you. You are but young still. Give Mrs. Paler ample
notice, three months, we believe, is the English usage—and
endeavour to part with her amicably. She must see that her situation
is beyond your strength."
I took the advice, and in June gave Mrs. Paler warning to leave,
having entered her house in September. She was angry, and affected
to believe I would not go. I respectfully asked her to put herself in
idea in my place, and candidly say whether or not the work was too
hard. She muttered something about "over-conscientiousness;" that
I should get along better without it. Nothing more was said; nothing
satisfactory decided, and the time went on again to the approach of
September. I wondered how I must set about looking out for another
asylum; I had no time to look out, no opportunity to go abroad. Mr.
Paler was in England.
"Miss Hereford, mamma told me to say that we shall be expected
in the drawing-room to-night; you, and I, and Harriet," observed
Kate Paler to me one hot summer's day. "The Gordons are coming
and the De Mellissies."
"What De Mellissies are those?" I inquired, the name striking upon
my ear with a thrill of remembrance.
 "What De Mellissies are those? why, the De Mellissies," returned
Kate, girl-fashion. "She is young and very pretty; I saw her when I
was out with mamma in the carriage the other day."
"Is she English or French?"
"English, I'll vow. No French tongue could speak English as she
does."
"When you answer in that free, abrupt manner, Kate, you greatly
displease me," I interposed. "It is most unladylike."
Kate laughed; said she was free-spoken by nature, and it was of
no use trying to be otherwise. By habit more than by nature, I told
her: and I waited with impatience for the evening.
It was Emily. I knew her at once. Gay-mannered, laughing, lovely
as ever, she came into the room on her husband's arm, wearing a
pink silk dress and wreath of roses. Alfred de Mellissie looked ill; at
least he was paler and thinner than in the old days at Nulle. She
either did not or would not remember me; as the evening drew on, I
felt sure that she did not, for she spoke cordially enough to me,
though as to an utter stranger. It happened that we were quite alone
once, in the recess of a window, and I interrupted what she was
saying about a song.
"Have you quite forgotten me, Madame de Mellissie?"
"Forgotten you!" she returned, with a quick glance. "I never knew
you, did I?"
"In the years gone by, when you were Miss Chandos. I am Anne
Hereford."
A puzzled gaze at me, and then she hid her face in her hands, its
penitent expression mixed with laughter. "Never say a word about
that naughty time, if you love me! everybody says it should be
buried five fathoms deep. I ought to have known you, though, for it
is the same gentle face; the sweet and steady eyes, with the long
eyelashes, and the honest good sense and the pretty smile. But you
have grown out of all knowledge. Not that you are much of a size
now. What an escapade that was! the staid Demoiselles Barlieu will
never get over it. I shall go and beg their pardon in person some
day. Were you shocked at it?"
"Yes. But has it brought you happiness?"
"Who talks of happiness at soirées? You must be as
unsophisticated as ever, Anne Hereford. Has that Johnstone left?"
"A long, long while ago. She was dismissed at the end of a few
months. The Miss Barlieus did not like her."
"I don't know who could like her. And so you are a governess?"
"Yes," I bravely avowed. "I have been nearly a year with the Miss
Palers."
"You must get leave to come and see me. Alfred, here's an old
schoolfellow of mine. I daresay you will remember her."
M. de Mellissie came at the call, and was talking to me for the rest
of the evening.
The great things that a night may bring forth! The sadness that
the rising of another sun may be bearing to us on its hot wings!
It was the morning following the soirée. I was in the schoolroom
with the girls, but quitted it for a minute to read a letter in peace
that arrived by the early post. It was written by Miss Barlieu. A very
kind letter, telling me to go back to them while I looked out for a
fresh situation, should I not get one before leaving Mrs. Paler.
Suddenly the door opened, and Mrs. Paler came in without any
ceremony of knocking, her face white, and an open letter in her
hand. She looked scared, fierce; agitation impeding her free
utterance.
"Here's news!" she brought out at length, her voice rising to a
scream; "here's news to come upon me like a thunderbolt! Does he
expect me to live through it?"
"Oh, Mrs. Paler, what has happened? You look ill and terrified. You
have had bad tidings! Will you not tell them to me?"
"What else have I come for but to tell you?" she retorted,
speaking in a tone that betrayed as much anger as distress. "I went
to the study after you, and frightened the girls; they were for
following me here, so I locked them in. I must tell some one, or my
feelings will burst bounds; they always were of a demonstrative
nature. Not like his, the sly, quiet fox!"
My fears flew to Mr. Paler. He had been in England some time
now, ever since the middle of May. Though I did not understand her
anger, or the last words.
"You have heard from Mr. Paler, madam!" I uttered. "Some harm
has happened to him!"
"Harm! yes, it has. Harm to me and my children, though, more
than to him. Miss Hereford, he has just gone and ruined himself."
"How?" I asked, feeling grieved and puzzled.
"It was always his mania, that turf-gambling, and as a young man
he got out of thousands at it. I thought how it would be—I declare I
did—when he became restless here in Paris, just before the Epsom
Meeting, and at last went off to it. 'You'll drop some hundreds over
it, if you do go,' I said to him. 'Not I,' was his retort, 'since I have
had children to drop hundreds over, I don't spare them for
racehorses.' A wicked, reckless man!"
"And has he—dropped the hundreds, madam?"
"Hundreds!" she shrieked; and then, looking covertly around the
roof, as if fearful others might be listening, she sunk her voice to a
whisper: "He has lost thirty thousand pounds."
"Oh!" I exclaimed, in my horror. Mrs. Paler wrung her hands.
"Thirty thousand pounds, every pound of it—and I hope remorse
will haunt him to his dying day! Epsom, Ascot, Goodwood—I know
not how many other courses he has visited this summer, and has
betted frantically at all. The mania was upon him again, and he
could not stop himself. He is lying ill now at Doncaster, at one of the
inns there, and his brother writes; he tells me they dare not conceal
the facts from me any longer."
"Shall you not go over to him, madam?"
"I go over to him!" she retorted; "I would not go to him if he were
dying. But that my children are his, I would never live with him
again; I would never notice him: I would get a divorce if practicable,
but for their sakes. You look shocked, Miss Hereford; but you, an
unmarried girl, cannot realize the blow in all its extent. Do you think
a man has any right wilfully to bring disgrace and misery upon his
wife and children?"
"Oh, madam—no!"
"It is my punishment come home to me," she wildly exclaimed.
"They told me how it would be, sooner or later, if I persisted in
marrying James Paler: but I would not listen to them. My mother
and sisters will say it serves me right."
I heard the children squealing and kicking at the schoolroom door,
and did not dare to go to them.
"It is next door to ruin," said Mrs. Paler; "it will take from us more
than half our income; and present debt and embarrassment it must
bring. Ah! see how some things—trifles—happen sometimes for the
best! I thought it a great misfortune to lose you, but I am glad of it
now, for I am sure I can no longer afford an expensive governess.
Nor many servants, either. Oh, woe's me!"
I stood looking at her distress with great pity, feeling that Mr. Paler
must be next kin to a madman. And yet I had liked him: he was
most affectionate to his children, and solicitous for the comfort of his
household. Mrs. Paler seemed to become suddenly awake to the
uproar. She darted to the schoolroom, scolded one, boxed another,
locked the door upon them again, and came back to me.
"I had better settle things with you at once, Miss Hereford. If I
take it in my head, I may go off to my family in England at a
minute's notice; there's no knowing. Your time here will expire in a
fortnight?"
"Yes."
"I had intended to offer an increased salary, if you would stay on
—but that's all out of the question now. I suppose you have no
settled plans; no fresh situation to go to?"
"Madam, it has not been in my power to look out for one."
 "True. Yet it is better that you should go. I don't know what may
become of us in future: where we shall live, or what we shall do—
perhaps go to some obscure place in Germany, or Scotland, or
Wales, and economize: anywhere, that it's cheap. I wonder that
such men, who deliberately bring ruin on their families, are
permitted to live! But now we must try and find you another
situation."
"Perhaps Madame de Mellissie may know of something: and I
think she would interest herself for me, if I knew how to see her."
"You can go and see her," replied Mrs. Paler, "you can go to-day,
and call upon her. My maid shall take you. Never mind the studies: I
feel as if I should not care if the girls never learnt anything again—
with this blow upon them."
I did not wait for a second permission: the thought that Emily de
Mellissie might help me to a fresh situation had been floating in my
mind all night. She was well-connected in England; she was in the
best society in Paris; and she was good-natured.
In the afternoon I proceeded to the hotel (as it was called) of old
Madame de Mellissie, for it was her house, and her son and
daughter-in-law lived with her. Emily was at home, surrounded by
morning callers, quite a crowd of them. She looked intensely
surprised at seeing me; was, or I fancied it, rather distant and
haughty in manner; and, pointing to a chair, desired me to wait. Did
she deem I had presumptuously intruded as one of those morning
callers? Very humbly I waited until the last had gone: schooling
myself to remember that I was but a poor governess, while she was
Madame Alfred de Mellissie, née Miss Chandos of Chandos.
"And so you have soon come to pay me a visit, Miss Hereford!"
"I have come as a petitioner, rather than as a visitor, Madame de
Mellissie. Can you spare me five minutes?"
"I can spare you ten if you like, now those loungers are gone."
I forthwith told my tale. That I was leaving Mrs. Paler's, where I
was overworked: that I had thought it possible she might know of
some situation open: if so, would she kindly recommend me?
 "The idea, Anne Hereford, of your coming to me upon such an
errand!" was her laughing answer. "As if I troubled myself about
vacant situations! There is a rumour current in Paris this morning
that James Paler has been idiot enough to go and ruin himself on
the turf. That he has lost a great deal of money is certain, for the
newspapers allude to it in a manner not to be mistaken. Thank
goodness, Alfred has no weakness that way, though he is empty-
headed enough. Is it not a dreadful life, that of a governess?"
"At Mrs. Paler's it has been one of incessant toil. I hope to go
where the duties will be lighter. It is not the life I like, or would have
chosen; but I must bend to circumstances."
"That's true enough. I will ask all my friends in Paris if they——by
the way," she abruptly broke off, speaking with slow deliberation, "I
wonder whether—if you should be found suitable—whether you
would like something else?"
I made no reply; only waited for her to explain herself.
"The case is this, Miss Hereford," she resumed, assuming a light
manner. "I thought of going to Chandos on a visit; my husband was
to have conducted me thither, but Madame de Mellissie has been
ailing, and Alfred says it would not do for him to leave her. This
morning we had a dispute over it. 'There's nothing much amiss with
her,' I said; 'were she in danger, it would be a different matter, but
it's quite unreasonable to keep me away from Chandos for nothing
but this.' Monsieur Alfred grew vexed, said he should not quit her,
and moreover, did not himself feel well enough to travel—for he has
a sort of French fever hanging over him. They are always getting it,
you know. I am sick of hearing one say to another, 'J'ai fièvre
aujourd'hui!' Then I said I should go without him: 'With great
pleasure,' he complacently replied, provided I would engage a lady
as companion, but he should not trust me alone. Complimentary to
my discretion, was it not?"
I could not deny it—in a certain sense.
"But the bargain was made; it was indeed. I am to look out for a
companion, and then I may be off the next hour to England;
destination Chandos. Would you like to take the place?"
A thousand thoughts flew over me at the abrupt question
crowding my mind, dyeing my cheeks. The prospect, at the first
glance, appeared like a haven of rest after Mrs. Paler's. But—what
would be my duties?—and was I, a comparative child, fit for the
post? Should I be deemed fit by Monsieur de Mellissie?
"What should I have to do?" I asked.
"Anything I please," she answered. "You must amuse me when I
am tired, read to me when I feel inclined to listen, play to me when
I wish, be ready to go out when I want you, give orders to my maid
for me, write my letters when I am too idle to do it, and post
yourself at my side to play propriety between this and Chandos.
Those are the onerous duties of a dame de compagnie, are they
not? but I have no experience in the matter. Could you undertake
them?"
She spoke all this curiously, in a haughty tone, but with a smile on
her face. I did not know how to take it. "Are you speaking seriously,
Madame de Mellissie?"
"Of course I am. Stay, though. About the payment? I could not
afford to give much, for my purse has a hole at both ends of it, and
I am dreadfully poor. I suppose you have had a high salary at Mrs.
Paler's?"
"Sixty guineas."
"Oh, don't talk of it!" she exclaimed, stopping her ears. "I wish I
could give it; but I never could squeeze out more than twenty. Anne,
I will make a bargain with you: go with me to Chandos, stay with me
during my visit there; it will not last above a week or two; and when
we return here, I will get you a more lucrative situation. For the time
you are with me, I will give you what I can afford, and of course pay
your travelling expenses!"
With the word. "Anne," she had gone back to the old familiar
manner of our school-days. I accepted the offer willingly, subject, of
course, to the approval of Monsieur de Mellissie; and feeling very
doubtful in my own mind whether it would be carried out. As to the
payment—what she said seemed reasonable enough, and money
wore but little value in my eyes: I had not then found out its uses.
Provided I had enough for my ordinary wants of dress, it was all I
cared for; and a large sum was due to me from Mrs. Paler.
Somewhat to my surprise, M. de Mellissie approved of me as his
wife's companion, paying me a compliment on the occasion. "You
are young, Mademoiselle Hereford, but I can see you are one fully to
be trusted: I confide my wife to you."
"I will do what I can, sir."
"You laugh at my saying that thing," he said, speaking in his
sometimes rather odd English. "You think my wife can better take
care of you, than you of her."
"I am younger than she is."
"That goes without telling, Mademoiselle. You look it. The case is
this," he added, in a confidential tone. "It is not that my wife wants
protection on her journey; she has her femme de chambre; but
because I do not think they would like to see her arrive alone at
Chandos. My lady is difficile."
The permission to depart accorded, Madame de Mellissie was all
impatience to set off. I bought a dress or two, but she would not
allow me time to get them made, and I had to take them unmade.
Though I was going to Chandos as a humble companion, I could not
forget that my birth would have entitled me to go as a visitor, and
wished to dress accordingly.
The foolish girl that I was! I spent my money down to one
Napoleon and some silver; it was not very much I had by me; and
then Mrs. Paler, to my intense consternation, told me it was not
convenient to pay me my salary.
She owed me thirty guineas. I had received the first thirty at the
termination of the half-year: it was all spent, including what I had
laid out now. I appealed to Mrs. Paler's good feeling, showing my
needy state. In return she appealed to mine.
"My dear Miss Hereford, I have not got it. Until remittances shall
reach me from Mr. Paler, I am very short. You do not require money
for your journey, Madame Alfred de Mellissie pays all that, and I will
remit it to you ere you have been many days at Chandos. You will
not, I am sure, object so far to oblige a poor distressed woman."
What answer could I give?
On a lovely September morning we started for Boulogne-sur-Mer,
Madame Alfred de Mellissie, I, and her maid Pauline. M. de Mellissie
saw us off at the station.
"I would have run down to Boulogne to put you on board the
boat, but that I do not feel well enough; my fever is very bad to-
day," he said to me and his wife. She took no notice of the words,
but I saw they were true: his pale thin face had a hectic red upon it,
his hand, meeting mine in the adieu, burnt me through my glove.
"Madame de Mellissie, your husband certainly has an attack of
fever," I said, as the train started.
"Ah, yes, no doubt; the French, as I previously observed, are
subject to it. But it never comes to anything."
 CHAPTER XI.
CHANDOS.

The station of Hetton, some fifty miles' journey from London on

the Great Western line, and two from Chandos, lay hot and bright in
the September sun. It was afternoon when we reached it. Madame
de Mellissie had preferred to stay a night in London, and go on the
next day at leisure. A handsome close carriage was in waiting
outside the station, its three attendants wearing the Chandos livery,
its panels bearing the arms of the Chandos family, surmounted by
the badge of England's baronetage, the bloody hand. The servants
lifted their hands to their hats, and respectfully welcomed Madame
de Mellissie.
"Is mamma well?" she inquired of them.
"Quite well, madam."
"And my brother? Why is he not here?"
"Mr. Chandos, madam, was obliged to attend a county meeting."
"Those ponderous county meetings!" she retorted. "And they
never do any good. Step in, Miss Hereford."
We were soon driving along. Pauline sat behind with one of the
footmen, the other remained to bring on the luggage. Madame de
Mellissie looked out on the points of road as we passed, with all the
glee of a child.
"This is my second visit only to Chandos since my marriage. For
two years mamma was implacable, and would not see me; but last
year she relented, and I came here for a little while. I don't believe,
though, mamma will ever forgive me in her heart. I am sorry for it
now."
"Sorry for having—having married as you did?"
"Ay, I am: Those rebellious marriages never bring luck. They can't,
you know; only, girls are so thoughtless and stupid. I made my own
bed, and must lie on it; it is not so bad as it might have been: but—
of course, all that's left is to make the best of it. Alfred says we
should get on better if we had children. I say we should not. And
there, in the distance, you see the chimneys of Chandos. Look,
Anne!"
She was wayward in her moods; wayward to me as to others.
Sometimes, during our past journey, she would be distantly polite,
calling me "Miss Hereford:" the next moment open and cordial as
ever she had been at school. That she had thrown herself away in a
worldly point of view, marrying as she did, was indisputable, and
Emily Chandos was not one to forget it.
Chandos was a long, low, red brick house, with gables and turrets
to its two end wings, and a small turret in the middle, which gave it
a somewhat gothic appearance. It was but two stories high, and
struck me as looking low, not elevated, perhaps partly from its
length. No steps ascended to the house, the lower rooms were on a
level with the ground outside. It was a sort of double house; the
servants' rooms, kitchens, and chambers, all looking to the back,
where there was a separate entrance. Extensive grounds lay around
it, but they were so crowded with trees, except just close to the
house, as to impart a weird-like, gloomy appearance; they
completely shut Chandos House from the view of the world beyond,
and the beyond world from the view of Chandos. A pretty trellised
portico was at the entrance; jessamine, roses, and clematis
entwined themselves round it, extending even to the windows on
either hand. Before the carriage had well stopped, a gentleman rode
up on horseback, followed by a groom. He threw himself from his
horse, and came to the carriage-door.
"Back just in time to receive you, Emily. How are you, my dear?"
 She jumped lightly from the carriage, and he was turning away
with her when he saw me. His look of intense surprise was curious
to behold, and he stopped in hesitation. Emily spoke: her tone a
slighting one, almost disparaging.
"It is only my companion. Would you believe it, Harry, Alfred took
a prudent fit, and would not suffer me to travel alone? So I engaged
Miss Hereford: she was in quest of a situation; and we knew each
other in days gone by."
He assisted me from the carriage. It was the same fine man I had
seen some years before at Mademoiselle Barlieu's; the same pale
countenance, with its delicate features and rather sad expression;
the same sweet voice. He then gave his arm to his sister, and I
followed them to the sitting-room. They called it the oak parlour; a
large, square room, somewhat dark, its colours harmoniously
blending, and its windows shaded with the trained clematis and
jessamine. It was the favourite sitting-room at Chandos. Other
reception-rooms there were: a gorgeous double drawing-room, a
well-stored library, a spacious dining-room; but the oak parlour was
the favourite. And none could wonder at it; for it was just one of
those seductive apartments that speak to the feelings of repose.
"Where's mamma?" exclaimed Emily, as we entered.
"Not far; she will be here directly, you may be sure," replied Mr.
Chandos. "Is this your first visit to our part of the country, Miss
Hereford?"
"Yes; I never was here before."
Now what was there in this reply to offend Madame de Mellissie?
or did she resent his speaking to me at all? She turned round,
haughty pride stamped on every line of her countenance, rebuke on
her tongue: though the rebuke lay in the tone, rather than in the
words.
"Miss Hereford! the gentleman to whom you speak is Mr.
Chandos."
Had I again omitted the sign of my dependent situation, the "sir?"
I, who had resolved, with my then burning face (burning again
now), never so to offend for the future—I supposed that that was
the meaning of Madame de Mellissie; I suppose so still, to this hour.
I had spoken as though I were the equal of Mr. Chandos: I must not
—I would not—so offend again.
"Emily, my love, you are welcome."
A little woman had entered the room, and was holding Madame de
Mellissie in her arms. It was Lady Chandos. She wore a small and
pretty widow's cap of net, a rich but soft black silk dress, and black
lace mittens. Her nose was sharp, and her small face had a
permanent redness, the result of disturbed health. She was not like
her daughter, not half so beautiful; and she was not like her
handsome son, unless it was in the subdued, sad expression. She
quite started back when her eyes fell on me, evidently not prepared
to see a stranger.
"Miss Hereford, mamma; a young lady whom I have engaged as
companion. Alfred would not suffer me to travel alone."
Lady Chandos turned to me with a pleasant smile, but it struck me
as being a forced one.
"I think you look more fit to take care of Miss Hereford, Emily,
than Miss Hereford of you," she said.
"I am the elder by some two or three years, if you mean that,
mamma. Oh! it was just a whim of my husband's."
More questioning on either side; just the information sought for
when relatives meet after a long absence. Emily answered carelessly
and lightly; and I sat behind, unnoticed.
Hill was called. Hill was still at Chandos, lady's-maid and
housekeeper, a confidential servant. She came forward, wearing a
dark brown gown and handsome black silk apron, her grey hair
banded under her close white lace cap. Lady Chandos spoke with
her in an undertone, most likely consulting what chamber I should
be placed in, for Hill turned her eyes upon me and looked cross.
A wide staircase, its balustrades of carved oak, gilded in places,
wound up to the rooms above. A gallery, lighted from above, ran
along this upper floor, from wing to wing, paintings lining it. It
seemed as if the wings had some time been added to the house, for
they were of a different style of architecture. A green-baize door
shut them out from the gallery. Beyond this was a narrow corridor,
and then a double door of stout oak, which formed the real entrance
to the wings: the same on both sides. What rooms might be within
them, I did not yet know. Each wing had a staircase of
communication between its upper and lower floors, and also a small
door of egress to the grounds on the sides of the house, where the
trees grew very thick. In the east wing (the house, you must
understand, facing the south), this lower outer door was kept locked
and barred—to all intents and purposes, closed up; in the west wing,
which was inhabited exclusively by Lady Chandos, the door was
simply locked, and could be opened inside at will; though no one
ever made use of it but herself, and she very rarely.
Several rooms opened from the gallery to the front—all of them
bed-chambers, except one: that was the library. The library was the
room next to the east wing. Opposite to it was a door opening to a
room that looked back, level with the north rooms in the east wing.
A similar room opened from the gallery at the other end. In fact, the
house was built in uniform—one end the same as the other. Between
the doors of these two rooms the wall of the gallery ran unbroken;
there was, in fact, no communication whatever, as regards the upper
rooms, between the back portion of the house and the front.
And now for the ground-floor. The portico was not in the middle of
the house, but near to the east wing; one room only, the large
dining-room, that seemed to be never used, lying between. The hall
was rather small, dark, and shut in, the oak parlour being on the left
hand as you entered. Two doors at the back of the hall led, the one
to the handsome staircase, the other to the kitchens and other
domestic rooms belonging to the household. A spacious corridor,
underneath the gallery above, branched off from the hall by means
of an open archway behind the oak parlour, and ran along the
house; and the various reception-rooms, all looking front, including
Mr. Chandos's private sitting-room, opened from it. A passage at the
other end of the corridor led to the rooms at the back, but it had
been closed up; and there was no communication whatever on this
lower floor with the wings. The doors in the hall, leading to the
stairs and to the servants' offices, as often as not stood open during
the day. Lady Chandos sat much in the west wing; she seemed to
like being alone. And I think that is all that need be said at present
in regard to the in-door features of the house. The description has
not been given unnecessarily.
Hill marshalled me up the staircase. It had been decided that I
was to have the "blue room." The stairs terminated in a wide
landing. The library and the east wing lay to the right, as we
ascended; the long gallery on the left. Hill passed two chamber-
doors, and opened a third, that of the blue room. It was as little
calculated for immediate occupation as any room can well be; the
whole of the furniture being covered up with clean sheets of linen,
except the blue silk window-hangings. Madame de Mellissie had the
room next to it, and I could hear her talking in it with her mother.
Hill surveyed matters, and gave a sort of grunt.
"Ugh! I thought the maids had uncovered this room yesterday: as
I've just told my lady. They must have hurried over their cleaning
pretty quick. Please to step this way, Miss. If you'll wait here a few
minutes, I'll have things arranged."
She went back along the gallery, opened the door of the first
bedroom on this side the staircase, and showed me in. It was a very
pretty room, not large; its hangings and curtains of delicate chintz,
lined with pale rose-colour, and its furniture not covered up, but as
evidently not in occupation. I wondered why they could not put me
in that. The window was wide open. I untied my bonnet and stood
there, Hill closing the door and going downstairs, no doubt to call up
the housemaids.
With the exception of the gravel drive below, and the green lawn
in front of it, its velvet softness dotted with the brightest flowers, the
place seemed to look upon nothing but trees, intersected with
gloomy walks. Trees of all sorts—low as dwarf shrubs, high as
towering poplars, dark green, light green, bright green. The walks
branched everywhere—one in particular, just opposite my window,
looked very gloomy, shaded as it was by dark pine-trees. I found
afterwards that it was called the Pine Walk. Why the place should
have struck upon me with a gloom, I can hardly tell; other people
might have seen nothing to justify the impression. "Chandos has
need to live in a world of its own," I thought, "for assuredly it is shut
in from all view of the outer world."
There arose a sound as of some one softly whistling. It came from
the adjacent window, one in the gallery, which must have been open
the same as mine. I did not like to lean forward and look. Another
moment, and the whistling ceased; some one else appeared to have
come up, and voices in conversation supervened. They were those
of Lady Chandos and her son, and I became an involuntary hearer of
what troubled me much.
"This is one of Emily's wild actions," said Lady Chandos. "She
knows quite enough of our unhappy secrets to be sure that a
stranger is not wanted at Chandos."
"Look for the most improbable thing in the world, mother, before
you look for discretion or thought in Emily," was the reply of Mr.
Chandos. "But this is but a young girl, unsuspicious naturally from
her age and sex: Emily might have introduced a more dangerous
inmate. And it may happen that——"
"I know what you would urge, Harry," interrupted the voice of
Lady Chandos. "But there's no certainty. There cannot be: and it is
most unfortunate that Emily should have brought her here. Every
night, night by night as they come round, I lie awake shivering; if
the wind does but move the trees, I start; if an owl shrieks forth its
dreary note, I almost shriek with it. You know what we have cause
to fear. And for a stranger to be sleeping in the house!"
"Yes, it is certainly unfortunate."
"It is more than that; it is dangerous. Harry, I have never, I hope,
done a discourteous thing, but it did occur to me to put this young
girl to sleep on the servants' side of the house. I think her being so
ladylike in appearance saved her from it, not my good manners. I
don't know what to be at."
 Mr. Chandos made no reply.
"I wish I had done it!" resumed Lady Chandos. "But there's
another thing—Emily might object: and to have any fuss would be
worse than all. Still, look at the risk—the stake! Is it too late, do you
think, Harry? Would it do to change her room now?"
"My dear mother, you are the best judge," observed Mr. Chandos.
"I should not change the room if I could possibly avoid it; the young
lady might consider it in the light of an indignity. Emily introduced
her in a slighting sort of manner; but her looks are refined, her
manners those of a gentlewoman."
"Yes, that's true."
"How long does Emily think of remaining?"
"She says two weeks. But she is uncertain as the wind. How could
she think of bringing a stranger?"
"Have you told her all?—why it is just now particularly
undesirable?"
"No. She never has been told. And I hope and trust she may be
gone again before—before trouble comes."
"Quite right; I should not tell her. Well, mother, as you ask my
opinion, I say things had better remain as arranged; let the young
lady occupy the blue room. How cross Hill looked over it!"
"Not without cause. I cannot think how Emily can have been so
senseless. It is just as if she had planned the annoyance—bringing
her here without writing! Had she written, I should have forbidden
it."
"Let us hope that nothing will happen."
"Harry, we cannot answer for it. Again, on Ethel's account a
stranger in the house is not desirable. Emily might have thought of
that."
The voices ceased; I suppose the speakers quitted the place; and
down I sat, overwhelmed with shame and consternation. To be
introduced in this unwelcome manner into a house, bringing
annoyance and discomfort to its inmates, seemed to me little less
than a crime; I could scarcely have felt more guilty had I committed
one.
And what was the mystery? That something or other was amiss in
the family was all too evident. "Have they got a ghost here?" I said
to myself, in peevishness. Involuntarily the long-past words of
Annette Barlieu flashed into my mind: and I had never thought of
them since they were spoken. "There is always a cloud hanging over
Chandos. They do not care to have a governess residing there:
Miladi said it to me." Then what was the cloud?—what was the fear?
Hill came in again, saying I was to keep the chintz-room. Lady
Chandos, in passing just now along the gallery to her own
apartments in the west wing, saw for the first time that the blue
room was not ready. So it was decided between her and Hill that I
should occupy the chintz one.
The luggage was brought up, and I began to dress for dinner. A
question occurred to me—are companions expected to dress, in the
wide sense of the term? I really did not know, in my inexperience.
My birth entitled me to do so; but did my position? A minute's
hesitation told me I was a guest at Chandos, treated and regarded
as one, and might appear accordingly. So I put on a pretty low blue
silk, with my necklace of real pearls, that had once been mamma's,
and the pale-blue enamelled bracelets with the pearl clasps. I had
been obliged to dress a good deal at Mrs. Paler's in the evening; and
—to confess the truth—I liked it.
I stood at the door, hesitating whether to go down, as one is apt
to do in a house, the ways of which are unfamiliar, when Mr.
Chandos, ready for dinner, came suddenly out of the room opposite
to the library, nearly opposite to mine, the one that I spoke of as
looking to the back of the house, and adjoining the back rooms of
the east wing. I concluded that it was his bed-chamber. He smiled at
me as he crossed to the stairs, but did not say anything. Directly
after, Emily de Mellissie appeared in the gallery, radiant in white silk,
with an apple-blush rose in her hair, and a diamond aigrette
embedded in it. They said she was full of whims—as I knew for
myself. How ardently I hoped that some whim would send her
speedily away from Chandos!
We went into the first drawing-room, one of the most beautiful
rooms I had ever seen, its fittings violet and gold. Lady Chandos was
there, and did not appear to have changed her dress. The dinner
was served in the oak-parlour; not once in a year did they use the
great dining-room. Lady Chandos kindly passed her arm through
mine; and Mr. Chandos brought in his sister.
It was a pleasant dinner, and a pleasant evening. Emily was on
her best behaviour, telling all manner of amusing anecdotes of Paris
life to her mother and brother, ignoring me. I listened, and was
spoken to by the others now and then. We did not quit the oak-
parlour. When dessert was taken, Hickens, the butler, removed it and
brought in tea. "After my snug sitting-room upstairs, the drawing-
room is so large," observed Lady Chandos to me, as if in apology; "I
like this parlour best."
Upon retiring to rest, a neat-looking servant with light hair, whose
name I found was Harriet, came to the chintz-room, and asked
whether she should do anything for me. She said she was one of the
housemaids—there were two besides herself, Lizzy Dene and Emma.
Altogether, including the coachman, a helper in the stables, and two
gardeners—all four of whom were out of doors, living half a mile
away—there were seventeen servants at Chandos. A large number,
as it seemed to me, considering the very little attendance that was
required of them. I told Harriet I had been accustomed to wait upon
myself, and she retired.
But I could not get to sleep. The conversation I had overheard
kept haunting me. I wondered what the mystery could be; I
wondered whether I should be disturbed in the night by noises, or
else. What uncanny doings could there be in the house?—what
unseemly inmates, rendering it inexpedient that a stranger should
share its hospitality? Was it really tenanted by ghosts?—or by
something worse? At any rate, they did not molest me, and my sleep
at last was tranquil.
 We went down the following morning at half-past eight; Emily in a
white dimity robe of no shape, but tied round the waist with a
scarlet cord, the effect altogether rather untidy; I in a mauve-
coloured muslin, with ribbons of the same shade; and found Lady
and Mr. Chandos waiting breakfast in the oak-parlour. The panels of
this room were of alternate white and carved oak, with a great deal
of gilding about both; it had a most unusual appearance; I had
never seen anything like it before. The ceiling was white, with gilt
scrolls round it, and cornices. The large chimney-glass was in a
carved oak frame, gilded in places to match the walls; the slanting
girandole opposite the window, reflecting the green grass and the
waving trees in its convex mirrored surface, had a similar frame. The
chandelier for the wax lights was of gilt, also the branches on the
mantelpiece, and those of the girandole. It was a pleasant room to
enter—as I thought that morning. The oak-brown silk curtains, with
their golden satin-wrought flowers, were drawn quite back from the
windows, which were thrown open to the lovely morning air; a
bright fire burnt in the grate opposite the door; the breakfast-table,
with its snow-white linen, its painted Worcester china, and its
glittering silver, was in the middle. Easy-chairs stood about the room,
a sofa against the wall—all covered to match the curtains—brown
and gold: a piano was there, a sideboard stood at the back,
underneath the reflective mirror; other chairs, tables, ornaments;
and the dark carpet was soft as the softest moss. Out of all order
though cavillers for severe taste might have called the room, I know
that it had an indescribable charm.
Lady Chandos, dressed just as she had been the previous day—
and I found it was her usual dress at all times—sat with her back to
the window, her son facing her, I and Emily on either side. Breakfast
was about half over when Hickens brought in some letters on a small
silver waiter, presenting them to Mr. Chandos. I was soon to learn
that all letters coming to the house, whether for servants or else,
were invariably handed first of all to Mr. Chandos. One of these was
directed to "Lady Chandos;" two to "Harry Chandos, Esquire;" the
fourth to "Mrs. Chandos." Mr. Chandos put his mother's letter on the
waiter again, and Hickens handed it to her. He then came back with
the waiter to his master, who placed the other letter upon it.
"For Mrs. Chandos." And Hickens went out with it.
Who was Mrs. Chandos? I should have liked to ask, but dared not.
"Do you mean to say that there is no letter for me, Harry?"
exclaimed Madame de Mellissie. "That's my punctual husband! He
said he should be quite certain to send me a letter to-day."
"The French letters often come in later, Emily," remarked her
brother.
He and Lady Chandos read their letters, Emily talked and laughed,
and the meal came to an end. At its conclusion Mr. Chandos offered
to go round the grounds with his sister.
"Yes, I'll go," she answered. "You can go also, Miss Hereford, if
you like. But we must get our bonnets and parasols, first, Harry."
My bonnet and parasol were soon got, and I stood at my bedroom
door, waiting for Emily. As she came down the gallery, the green-
baize door on my right, leading to the east wing, opened, and a
middle-aged lady appeared at it. Madame de Mellissie advanced and
cordially saluted her.
"I should have paid you a visit yesterday, Mrs. Freeman, but that I
heard Mrs. Chandos was ill."
"You are very kind, madam," was the lady's reply. "Mrs. Chandos
was exceedingly unwell yesterday, but she is better to-day. She——"
Mrs. Freeman was interrupted. A lovely-looking girl—girl she
looked, though she may have been seven or eight-and-twenty—
appeared at the door of one of the rooms in the wing. Her dress was
white; she wore a beautiful little head-dress of lace and lavender
ribbons, and she came forward, smiling.
"I heard you had arrived, Emily dear, and should have joined you
all yesterday, but I was so poorly," she said, clasping Madame de
Mellissie's hand. "How well you look!"
"And you look well also," replied Emily. "We must never judge you
by your looks, Mrs. Chandos."