Convergent and divergent validity

http://dissertation.laerd.com/convergent-and-divergent-validity-p2.php
Convergent validity and divergent validity are ways to assess the construct validity of a
measurement procedure (Campbell & Fiske, 1959). If you are unsure what construct validity
is, we recommend you first read: Construct validity. Convergent validity helps to establish
construct validity when you use two different measurement procedures and research
methods (e.g., participant observation and a survey) in your dissertation to collect data
about a construct (e.g., anger, depression, motivation, task performance). Divergent
validity helps to establish construct validity by demonstrating that the construct you are
interested in (e.g., anger) is different from other constructs that might be present in your
study (e.g., depression). To assess construct validity in your dissertation, you should first
establish convergent validity, before testing for divergent validity. In this article, we explain
what convergent and divergent validity are, providing some examples.
What is convergent validity?
What is divergent validity?
What is convergent validity?
Convergent validity helps to establish construct validity when you use two
different measurement procedures and research methods (e.g., participant observation and
a survey) in your dissertation to collect data about a construct (e.g., anger, depression,
motivation, task performance). The extent to which convergent validity has been
demonstrated is establish by thestrength of the relationship between the scores that are
obtained from the two different measurement procedures and research methods that you
have used to collect data about the construct you are interested in. The idea is that if these
scores converge,despite the fact that we use two different measurement procedures and
research methods, we must be measuring the same construct.
We use the words, despite the fact, because it can be difficult enough in research to
create one reliable operational definitionfor a construct; that is, a single reliable way
of measuring a particular construct. It's one thing to suggest measuring the
construct height using centimetres, or a person's weight using kilograms, but these
are operational definitions of constructs that are quite obvious, where it is easy to come up
with a single operational definition. It is far more challenging to create reliable operational
definitions for constructs like anger, depression, motivation, and task performance, let
alone multipleoperational definitions [see the article on Constructs in quantitative research].
However, in order to establish convergent validity, we must come up with two operational
definitions of the construct we are interested in. We have to come up with twooperational
definitions because we are using two different measurement procedures (e.g., with
participant observation and a survey as the research methods). Each of these measurement
procedures will require a different operational definition. Let's look at an example:
Study #1
Construct #1 = Sleep quality

Imagine that we are interested in studying the relationship between fitness
level and sleep quality; that is, the impact that exercise has on how well
people sleep. For the purpose of this example, let's focus on the scores on
the dependent variable, which is sleep quality (i.e., sleep quality is
the construct of interest). When participants in the study wake up in the
morning, they record their sleep quality using a self-completed survey (i.e.,
they fill in a questionnaire). This gives us insight into how well the participants
felt they slept. However, is this a reliable measurement procedure to measure
the construct, sleep quality? Let's imagine that we are simply unsure because
sometimes self-completed measurement procedures can be prone to
certain biases. Therefore, we also observe the participants whilst they are
sleeping using a video camera to monitor their sleeping patterns. When
making the observations, we score the participants' sleep quality. We hope
that by using two different research methods to assess sleep quality, we will
have a more reliable measurement procedure for the construct we are
interested in.

This leaves us with two different sets of scores from the two different
measurement procedures used under the two research methods (i.e.,
the scores from the survey and the scores from the participant observation).
We will have started to demonstrate convergent validity if there is a strong
relationship between the two scores (i.e., the scores from the measurement
procedures used under the two different research methods). Such a strong
relationship, which helps to demonstrate convergent validity, is an important
step in assessing construct validity; that is, we can be more confident that the
measurement procedures that we are using to measure sleep quality are
a valid measure of theconstruct, sleep quality.
In order to establish convergent validity, the strength of the relationship between the scores
from the two different measurement procedures, from the two different methods, is
assessed. This is usually achieved by calculating a correlation between the two scores.
NOTE: Convergent validity is not the same as concurrent validity, which we discuss in more
detail in the article: Concurrent validity. However, the distinction is quite straightforward.
Both convergent and concurrent validity are ways of assessing construct validity by
examining the strength of the relationship between the scores from two different
measurement procedures. However, whilst concurrent validity compared
a new measurement procedure with a well-established measurement
procedure, both measurement procedures are new when testing for convergent validity.
Therefore, if one of the measurement procedures you are using to establish construct
validity is well-established, you should read the article: Concurrent validity.

What is divergent validity?
Divergent validity helps to establish construct validity by demonstrating that the construct
you are interested in (e.g., anger) is different from other constructs that might be present
in your study (e.g., depression). To assess construct validity in your dissertation, you should
first establish convergent validity, before testing for divergent validity.
Divergent validity is important because it is common to come up with an operational
definition for a construct that actually measures more than one construct. Unfortunately, we
are typically not aware that this has happened; after all, if we had, we wouldn't have made
the mistake in the first place; that is, we would have come up with a more
reliable operational definition. For example, we think that the questions we ask in a survey
about the construct, anger, only measure anger, when in fact they also measure another
construct, depression. In order to establish that the scores we obtained when collecting data
reflect anger and not depression, we need to test for the divergent validity of the
measurement procedures we used to capture angerand depression. To do this, we will
have two different measurement procedures and research methods to measure both
constructs we are examining. This could mean that we have a total of four measurement
procedures, but often you will have used the same research method to collect data for both
constructs (e.g., you used participant observation to measure
bothanger and depression amongst your sample, following this up with a survey, which
included questions also measuring both anger and depression).
The extent to which divergent validity has been demonstrated is establish by the strength of
the relationship between thescores that are obtained from the two different measurement
procedures and research methods that you have used to collect data about the two
constructs you are interested in. Unlike convergent validity, where we are interested in the
extent to which the scores converge (i.e., we want to see a strong relationship between
the two scores on the same construct), with divergent validity, we are interested in the
extent to which the scores diverge (i.e., we want to see little or no relationship between
thetwo scores from the two constructs). This is a two-step process:
1. Establish convergent validity: A strong relationship should be
established between the two scores for each of the two constructs (e.g., a
strong relationship for anger and a strong relationship for depression).
2. Establish divergent validity: Little or no relationship should be found
between the two scores between the two constructs(e.g., little or no
relationship between anger and depression) when comparing the same
methods used to collect the data (e.g., comparing anger and depression
from the observational scores, and comparing anger and depression from
the survey scores).
Let's look at an example:
Study #2
Construct #1 = Sleep quality
Construct #2 = Sleep quantity
Note: Quality vs. Quantity of Sleep

Let's imagine that in Study #1 we were able to establish a strong relationship
between the two sets of scores from the two different measurement
procedures under the two research methods (i.e., the scores from the survey
and the scores from the participant observation); in other words, we started to
establish convergent validity for the construct, sleep quality. However, now
that we look back at Study #1, we are concerned that we included sleep
quantity within the same set of measures (e.g., the questions in the survey)
that we used when measuring sleep quality. We say that we are concerned
about including these measures within the same measurement procedure
because we are unsure whethersleep quality and sleep quantity are part of
the same construct or are two different constructs (i.e., let's imagine that no
previous studies are able to answer this question for us). Now if sleep quality
and sleep quantity are two different constructs, but we measured them as if
they were the same construct, we have introduced a confounding variable that
will inevitably reduce the internal validity of our study [see the
articles: Extraneous and confounding variables andInternal validity].
Therefore, we decided to examine whether sleep quality and sleep quantity are
different constructs.

To achieve this, we use the same research methods as in Study #1; that is,
we ask participants to complete a survey, as well as observing participants
whilst sleeping. However, the survey contains (a) questions that measure
sleep quality and (b) questions that measure sleep quantity. Similarly, when
we observe participants, we record scores separately for(a) sleep quality
and (b) sleep quantity. In order to assess whether the two constructs (i.e.,
sleep quality and sleep quantity) are different, we first need to find that both
constructs have convergent validity. Therefore, there should be a strong
relationship between the survey scores and observational scores for (a) sleep
quality and (b) sleep quantity. Next, we need to find that these two constructs
are distinct; that is, that we have divergent validity. Therefore, there should
be little or no relationship between (a) the survey scores for sleep quality and
the survey scores for sleep quantity and (b) the observational scores for sleep
quality and the observational scores for sleep quantity. If this is the case, we
can be more confident that sleep quality and sleep quantity are, in fact, two
separate constructs. Since we had to establish convergent validity before we
could establish divergent validity, we can also be satisfied that we have
created two valid measurement procedures for sleep quality and sleep quantity
(i.e., a survey and observational measurement procedure for sleep quality,
and a survey and observational measurement procedure for sleep quantity).
Construct validity can start to be established when you:
1. Find that the scores that are obtained from the measurement
procedures you used from two different methods to assess
theconstruct you are interested in are strongly related; that is, the scores
converge, suggesting that both measurement procedures reflect the
construct you are interested in, establishing convergent validity.
2. Find that the scores obtained for the two constructs you are interested
in diverge (i.e., are unrelated); that is, there is little or no relationship
between the scores for the two constructs when comparing these scores
using the same methods. This establishes divergent validity.
We say that construct validity can start to be established when both convergent and
divergent validity are established because construct validity is something that is built over
time. No single study can establish construct validity [see the article: Construct validity].