MARKETING ENGINEERING FOR EXCEL

TUTORIAL

VERSION v130522

Tutorial

Customer Choice (Logit)

Marketing Engineering for Excel is a Microsoft Excel add-in. The software runs from
within Microsoft Excel and only with data contained in an Excel spreadsheet.
After installing the software, simply open Microsoft Excel. A new menu appears,
called MEXL. This tutorial refers to the MEXL/Customer Choice (Logit)
submenu.

Overview
The customer choice (logit) model is an individual-level response model that
helps analyze and explain the choices individual customers make in a market.
The customer choice model helps firms understand the extent to which factors
such as the price of a brand or its ease of installation influence a customer's
choice. A brand's purchase probability at the individual level can be
aggregated to determine the brand's market share at the market level.
Firms also can use customer choice analysis to develop marketing programs
tailored to specific market segments, or even to individual customers.
Further, if a company has purchase data about its products versus those of its
competitors (product choice data), as well as some observed independent
variables (e.g., gender, price, promotion), it can use customer choice modeling
to answer such questions as:

Does a customers gender influence his or her purchase decision

regarding our product(s)?

Do competitors promotions affect the purchase of our product(s)?

How do our promotions affect our sales rates?

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

1/15

Getting Started
Customer choice analysis allows you to use your own data directly or to use a
preformatted template.
Because customer choice models require a specific data format, users with
their own data should to review the preformatted template to learn about the
appropriate structure.

This section explains how to create an easy-to-use template to enter your own data.
If you want to run customer choice analysis immediately, open the example file
OfficeStar Data (Customer Choice).xls and jump to Step 2: Running analysis. By
default, the example files install in My Documents/My Marketing Engineering/.

Step 1

Creating a template
Using the interactive assistant
In Excel, if you click on MEXL CUSTOMER CHOICE (LOGIT) CREATE TEMPLATE, a
dialog box will appear. This box represents the first step in creating a template
for running the customer choice model. The first dialog box prompts you to
use the interactive assistant.
Marketing Engineering for Excel models allow you to use the interactive
assistant, which will prompt you to enter the parameters required by the
model and build a template spreadsheet for you to enter the required data.
This generated template includes pre-selected cell ranges that correspond to
the parameters you enter.
We recommend that you use Marketing
Engineering for Excel templates to simplify subsequent analyses.
Expert users who are familiar with Marketing Engineering for Excel models and
data requirements may prefer to input data directly in an unformatted
spreadsheet. Such users should begin with the interactive assistant to become
familiar with the data format that Marketing Engineering for Excel expects.
Unless you are already familiar with the methodology, you should select yes
in response to the box that asks whether you want to use the interactive
assistant.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

2/15

With the interactive assistant, the following dialog box allows you to create a
template to run the customer choice software. All customer choice models
require at least one case per respondent (customer) and at least one
independent variable.

The options are as follows:

Number of respondents: The total number of respondents (customers)

in your study.

Number of cases per respondent: The number of observations per

respondent (one or more than one purchase event).

Number of independent variables: The number of independent

variables you collected or observed during the study, such as respondent
gender, product on sale, and so forth.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

3/15

After completing the dialog box, click Next to move to the subsequent dialog
box and define the customer choice template further.

There are two types of data that can be used in customer choice analysis:

One alternative (Binary choice model). This data format is suitable when
customers have two choice alternatives, such as a choice between buy
and dont buy. That is, it requires a yes-no decision process.

Multiple alternatives (Multinomial choice model). This data format

considers customer choices across a subset of related competitors, such as
brands A, B, and C. Therefore, it requires a one-out-of-many decision
process.

If you select the multiple alternatives option, you will activate the lower
portion of the dialog box, which allows you to enter the labels you want to use
to describe the various alternatives. In the box labeled Enter new Alternative
Label, enter the label for the first alternative, then click the Add button. The
alternative label appears in the list, as shown below:

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

4/15

Repeat the process of adding options until you have entered all the desired
labels. You may use the Up, Down, or Delete buttons to modify your list by
simply selecting any item within the list.
If you specified the Single alternative model, the labels will automatically be
designated as Response (when the alternative is chosen) and Dummy
(when the alternative is not chosen).

If your data require a no choice option, you may specify it yourself. If you do not
specify a label for the no-choice option, the software will add this option
automatically during the analysis.
For example, if customers must choose among three brands, A, B, and C, and some
choose none of the options, the choice process actually involves four options (A, B,
C, and none). The software identifies this issue and adds the fourth option (called a
dummy) automatically. Alternatively, you may specify the four options manually.

After entering all the alternatives, click OK to generate the data collection
template, as shown below:

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

5/15

Not using the interactive assistant

If you decide not to use the interactive assistant, the following dialog box
appears:

You should use the Marketing Engineering for Excel templates whenever
possible; for example, the preceding dialog box allows you to specify the
parameters to use to build your template, but you must fill in the resultant
generated spreadsheet manually.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

6/15

Step 2

Entering your data

In this tutorial, we use the example file OfficeStar Data (Customer Choice).xls,
which by default appears in My Documents/My Marketing Engineering/.
To view a proper data format, open that spreadsheet in Excel. A snapshot is
reproduced below.

For each respondent, a customer choice spreadsheet contains:

Alternative names for each product in the study. If the spreadsheet was
generated using the single alternative option, this column is not present.

Cases. The same customer might have been exposed to the same choice
situation, repeatedly, and these repetitions assist model estimation. If the
spreadsheet was generated using the one case per respondent option, this
column is not present.

Choice variables. The alternative the customer chose receives a value of

1; all other alternatives receive values of 0. Only one 1 can exist per
case. If all options are set to 0, a none-of-the-above option (dummy) is
automatically inserted during the analysis.

Independent variables. A column represents each variable specified for

the study, and all independent variables should have consistent value
ranges. That is, the data for a single independent variable should be scaled
within the same range. Independent variables can take on discrete values
if they are appropriately specified using dummy-variable coding.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

7/15

Step 3

Running analyses
After you enter your data in an Excel spreadsheet with the appropriate format,
click on MEXL CUSTOMER CHOICE (LOGIT) RUN ANALYSIS. The dialog box that
appears allows you to set the options to perform the customer choice analysis
of your data.
The four sections in this dialog box control the customer choice analysis.

Data
In this portion of the dialog box, you must indicate which data are present in
the spreadsheet. If you have used the interactive assistant, the software
automatically clicks the appropriate check boxes. If you are using a custom
spreadsheet, you will need to select the appropriate check boxes to indicate
which data are present.

Generate Alternative-Specific Constants. In general, it is a good idea

to include alternative-specific constants in the model. If you check this
option, the program will automatically generate the dummy variables for
alternative-specific constants. For estimation purposes, the last listed
alternative in the data is set to the base value of 0. The alternativespecific constants for the other alternatives are estimated with respect to
this base.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

8/15

Significance level. Expressed in percentage terms, this option refers to

the significance level of the statistical tests. The software uses this value
to identify statistically significant coefficients from the analysis.

Perform analysis on holdout sample. This option allows you to specify

a number between 0 and 100 (0 is the default) to represent the
percentage of the data you want to use as a holdout sample to assess the
predictive validity of the model. If you choose this option, the customer
choice analysis selects the closest feasible sample with a size equal to, or
smaller than, the percentage you specify. For example, if you have 110
customers (cases) with one observation per customer and you specify
15%, the program sets aside the last 16 observations in the data file as
the holdout sample and estimate the model using the first 94 customers.

Latent Class Options

In this portion of the dialog box, you may indicate the maximum number of
segments which MEXL should consider. The default number of segments is set
to 1 which is the most common option. If you have reason to believe your data
may have multiple latent segments, you may select a number greater than
1. In the step that follows, you will be able to review the attributes of each
segment and decide how many segments to include in the analysis.
Note that for the latent class analysis to be executed, you need a sufficient
number of respondents, so that even small segments, if they occur (say a
segment that is about 5% of your sample), will still have a reasonable number
of respondents in that segment (say 20 or more respondents). Note: The
OfficeStar example used for this tutorial does not include the data necessary to
use the Latent Class option.
Advanced Menu
When using the Latent Class option, it may be necessary to further control the
evaluation criteria from the normal software defaults. Clicking the Advanced
button will produce the following expanded dialog box:

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

9/15

Latent Class Estimation Options

Latent Class analysis is implemented in MEXL via the EM algorithm. EM
(Expectation Maximization) is an iterative technique in which, in each iteration,
the procedure predicts the unobserved latent classes (in the E-step) and based
on these predictions re-estimates the parameters (in the M-step). The controls
below allow user to influence the execution of the program. Typically the
tolerance levels can be 0.001. The program uses several criteria to evaluate
whether to terminate, including how fast the likelihood function is converging,
and how fast the coefficients of the model are converging.
E steps: Specifies the number of expectation steps.
M steps: Specifies the number of maximization steps.
Log Likelihood, Gradient, Parameter and Objective Tolerances: These
parameters control the tolerances within the analysis. If a data set is not
converging, the user should increase the tolerance levels to gain better insight
about the data. Note that decreasing tolerance levels will increase the required
processing time.
Multiple Runs?: In many cases, it will be necessary to make multiple runs of
the same data to determine whether a true latent class exists. This parameter
will allow you to re-run the model n times, and select the best results from
those runs.
After selecting the desired options, click Next. The software will ask you to
select the data on which to perform the analysis.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

10/15

Optional Screen if Latent Class Option was Selected

If you changed the number of segments in Latent Class Options to greater
than 1, the following dialog box will appear. With this dialog box, you can
review the diagnostic properties associated with each number of segments, up
to the maximum number of segments you specified. Use the drop down arrow
to select segment scheme to be displayed. In the two sections of the screen,
use the scroll arrows to view all of the coefficient estimates for each segment.
The bottom part of this screen allows you to select how many segments you
wish to retain for analysis. Note: Latent Class will identify potential clusters
(segments). Numerous runs may be required before one can establish clusters
of importance.

Step 4

Interpreting the results

The generated workbook contains several worksheets that describe the
analysis.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

11/15

Variable Averages
This portion of the spreadsheet presents the means of each independent
variable for the group of respondents.
Variable Averages for Chosen Alternatives
This portion of the spreadsheet presents the means of each independent
variable where that alternative was the chosen alternative.
Confusion Matrix on Estimation Sample
The confusion matrix portion of the spreadsheet includes two matrices:
numerical counts and percentages for the same data.
The diagonal of both matrices appears bolded, to highlight the convergence of
the observed and predicted data. A high value or percentage at this diagonal
intersection represents a high correlation between observed and predicted
behavior.

Coefficient Estimates
This portion of the spreadsheet shows the coefficients associated with each
independent variable. Coefficients that are significantly different from 0 (at the
level of significance specified during the Run Analysis set up) appear in bold;
they are in green when positive (positive influence on the likelihood of choice)
and in red when negative.
In the OfficeStar example, past purchases has a positive influence on the
likelihood of choice (a customer is more likely to return to a store that he or
she has patronized in the past), whereas expensive has a negative impact
(people who perceive a store as expensive avoid it).
Elasticities
This portion reveals the elasticities of each independent variable in the study.
The matrix provides information about the elasticity of impact that each
variable has on choice shares. For a 1% change in an attribute in a given row,
the elasticity equals the percentage change expected in the response to that
alternative. Large values mean that changes in the variable represented in
that row have significant effects on the choice probabilities of the choice
alternatives in a specific column. The bolded cells on the diagonal indicate the
effect of a product change on that product, whereas the off-diagonal cells
reflect the effects of the same product change on a competitors product.
CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

12/15

In the OfficeStar example, service elasticity (.14) is rather small (compared

with other elasticities that appear in the diagonal of each elasticity matrix), so
improving perceptions of service quality by 1% would have a limited impact on
the likelihood of customer purchase from OfficeStar (i.e., .14%). However, the
greater service elasticity for a competitor, Office Equipment (.42), suggests
that this competitor would benefit much more than OfficeStar if it managed to
improve customers perceptions of its service quality. The off-diagonal
elements represent cross-elasticities.
The off-diagonal elements reveal the influence on competitors created by
changes in customer perceptions.

Estimated Choice Probabilities

This portion of the above spreadsheet shows the probabilities of segment
membership in each segment, the choice alternative that the model predicts
will be chosen by that respondent, and the observed choice for each
respondent in your study. The matrix will show all respondents unless you
selected the Perform analysis on holdout sample option. In this case, this
matrix shows only the results for the holdout sample.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

13/15

Step 4a Interpreting the results when using Latent

Class option
When using the Latent Class option, several screens will differ from those
produced when not using the Latent Class option. Notably, a worksheet will be
produced for each segment analyzed, as shown below.
Note: The OfficeStar example used for this tutorial does not include the data
necessary to use the Latent Class option.

The Hold-out worksheet will show the details of the analysis for the hold-out
sample.

Latent Class analysis also produces an additional worksheet, Segment

Membership, as shown below. The segment membership will show the
probability of each respondent belonging to each segment, along with the
predicted segment to which the respondent will belong (this is the segment to
which a respondent has the highest probability of belonging).
CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

14/15

If you requested analysis on a holdout sample, the Segment Membership tab

will show the results for the holdout population only, as shown below.

CUSTOMER CHOICE (LOGIT) TUTORIAL V130522

15/15