Appendix A: Additional Topics A.1 Categorical Platform (Optional)

Preview:

Citation preview

Appendix A: Additional Topics

A.1 Categorical Platform (Optional)A.1 Categorical Platform (Optional)

Objectives Identify specific cases for which the Categorical

platform was designed. Summarize complex categorical data with

the Categorical platform.

2

Introduction The purpose of the Categorical platform is to tabulate

and summarize categorical response data.

– Test statistics are available, too. The main advantage of the Categorical platform

is that it recognizes common formats for complex data collection.

– Surveys

– Clinical trials

– Quality assurance This platform reduces or eliminates the need

to reshape the data before analysis.

3

Analysis Roles Nine specific roles for the response (Y) are included.

– These roles address the complex data formats (next).

The optional X role provides for grouping variables.– The levels define samples or sample groups.

The optional Sample Size role is used for calculation of the rate of occurrence.

The optional ID role is used to collect multiple responses that appear on separate rows.

4

Response Role 1: Separate Responses The responses occur individually, in separate

columns. Each response might have different categories. A separate analysis is performed for each response.

5

ID School Grade Subject

GW PS101 6 History

LP PS107 11 Chemistry

WL PS104 2 Knitting

Response Role 2: Aligned Responses All of the responses use the same categories.

6

ID Proposal A Proposal B Proposal C

GW Pass Pass Fail

LP Pass Fail Pass

WL Fail Fail Fail

Response Role 3: Repeated Measures All of the responses use the same categories,

but they are measured more than once. This role provides an optional transition report

to determine whether categories change over time.

7

ID 2009 2010 2011

GW History English History

LP Physics Chemistry Chemistry

WL Knitting Knitting Knitting

Response Role 4: Rater Agreement All of the responses use the same categories to rate

the same item.

8

Item GW LP WL

History Like Like Dislike

Chemistry Dislike Like Dislike

Knitting Like Like Dislike

Response Role 5: Multiple Response All of the responses use the same categories that are

entered into separate columns, but treated as a single grouped response.

9

ID Subject 1 Subject 2 Subject 3

GW History Literature

LP Chemistry Physics Biology

WL Knitting

Response Role 6: Multiple Response by ID All of the responses use the same categories that

are entered into one column and one or more rows, but treated as a single grouped response.

Must use ID role to collect responses.

10

ID Subject

GW History

GW Literature

LP Chemistry

LP Physics

LP Biology

WL Knitting

Response Role 7: Multiple Delimited All of the responses use the same categories that

are entered into one column and one row, separated by commas.

11

ID Subject

GW History, Literature

LP Chemistry, Physics, Biology

WL Knitting

Response Role 8: Indicator Group The responses are binary across multiple columns

in a related group (ID).

12

ID Fall Winter Spring

GW Y Y Y

LP Y Y Y

WL N N Y

Response Role 9: Response Frequencies The responses are counts across multiple columns

in a related group (ID).

13

ID Fall Winter Spring

GW 3 2 3

LP 4 5 3

WL 0 0 1

Unique Occurrences This option enables you to count a response level just

once when it is duplicated for the same ID.

14

Grouping Options There are three options that control the results for your

grouping variables (X):– Combinations: This option results in frequency

reports for combinations of the samples.– Each Individually: This option results in frequency

reports separately for each samples.– Both: This option results in frequency reports both

ways described above.

15

Report: Frequency The Frequency report presents a tabulation

of the counts for each category and the total counts (Responses) and total units (Cases).

Grouping variables (X) produce a stratified tabulation.

16

Report: Share of Responses The Share of Responses report presents a tabulation

of the proportion of the total counts (Responses) for each category.

Grouping variables (X) produce a stratified tabulation.

17

Report: Rate per Case The Rate Per Case report presents a tabulation of the

proportion of the total units (Cases) for each category. Grouping variables (X) produce a stratified tabulation.

18

Report Format The reports are formatted as simple or stratified

tables, one each for the frequency, share, and rate. The Crosstab format optionally collects all three values

into one cell. The Table and Crosstab formats can be transposed.

19

Chart: Share The Share Chart presents a mosaic plot of the

proportion of the total counts (Responses) for each category.

Grouping variables (X) produce a stratified tabulation.

20

Chart: Frequency This chart presents a bar chart of the frequency table. Grouping variables (X) produce a stratified tabulation. This chart is optional. It is not presented by default.

21

Available Statistics Test Response Homogeneity Test Each Response

– Relative Risk– Conditional Association

Agreement Statistic Transition Report

22

Test Response Homogeneity Determine whether the proportions or probabilities

are the same across all samples. This marginal homogeneity test of independence

is based on the Pearson chi-square test and the likelihood ratio chi-square test.

Typically used when there is one response variable and one explanatory variable.– Multiple explanatory variables are treated

as one variable.

23

Test Each Response Determine whether the rates for each category

are the same across all samples. This test is based on Poisson regression.

– Model each response category separately.– The test is a likelihood ratio test.

24

Relative Risk (Optional) The risk (probability) of each category is computed

for each sample. The risks are compared across samples. This statistic requires the Unique occurrences within

ID option in the launch dialog.

25

Conditional Association (Optional) The probability of each category is computed, given

one of the other categories. This statistic requires the Unique occurrences within

ID option in the launch dialog.

26

Agreement Statistic Determine whether the ratings from each rater agree.

– This requires Rater Agreement response role.– The responses must use the same categories.– This is stricter than association.– The agreement test is based on Cohen’s kappa

statistic. Determine whether the lack of agreement

is symmetrical.– The symmetry test is based on Bowker

and McNemar statistic.

27

Transition Report Determine whether the frequencies of categories have

changed over time.– This requires the Repeated Measures response

role. This test is based on the counts and the rates

of the transitions.

28

Financial Advisor Survey Example A small firm that provides financial management

services sends a survey to customers to assess satisfaction.– Service Quality: rated Low, Medium, or High– Responsiveness: rated Low, Medium, or High– Years as Client– Type of Account: Individual or Business

Analyze Service Quality and Responsiveness over the samples of Years as Client and Type of Account.

29

Transform Samples Convert the continuous explanatory variable Years

as Client to a categorical variable, Client Retention.– All survey respondents are current clients.– Client Retention: New (1 to 3), Steady (4 to 8),

or Loyal (>8)

30

This demonstration illustrates the concepts discussed previously.

Categorical Platform

Exercise

This exercise reinforces the concepts discussed previously.

Recommended