# Version 19.5 Enhancements

Two new procedures have been added:

• Ordinal Regression - construction of regression models when the dependent variable Y consists of ordered categories, such as a Likert scale.
• Multinomial Logistic Regression - construction of regression models when the dependent variable consists of categories with no natural ordering, or when the proportional odds assumption of ordinal regression does not hold.

Significant improvements have also been made to several procedures throughout the program. These include:

Ordinal Regression
A new procedure has been added to fit regression models when the dependent variable is ordinal, as in the data below:

Typical output shows how the probability of each level of the dependent variable changes based on the values of one or more independent variables, which may be either quantitative and categorical.

Model fitting options include stepwise variable selection:

Multinomial Logistic Regression
A second new procedure fits regression models for situations where the dependent variable is categorical but not ordinal, as in the data below:

For such data, the models show how the probability of each level of the dependent variable varies as a function of the independent variables.

One Variable Analysis and One Sample Analysis SnapStat
The data input dialog box for these procedures has been expanded to allow for the input of data based on values and frequencies. For example, consider the data below showing how many tries were necessary to guess the word presented by the New York Times Wordle puzzle for a sample of 300 games:

 Tries Games 1 0 2 5 3 43 4 125 5 93 6 32 7 2

To calculate statistics such as the average number of tries, the data input dialog box can be completed as follows:

This permits a more intuitive analysis than having to enter REP(Tries,Games) in the Data field as well as improving the labeling in tables and graphs.

Oneway ANOVA, Multiple Sample Comparison, Multifactor ANOVA and General Linear Models

The table generated  by the Multiple Range Analysis now creates columns of letters rather than X's to indicate members of each homogenous group:

The letters may also be added to the Means Plot to signify the homogenous groups into which the level means have been grouped. For example, the plot below shows that materials 1, 2 and 4 all belong to different groups. No conclusion has been reached about material 3, which may below to either group A or group B.

Barchart, Multiple Barchart, Tabulation, Crosstabulation, Pareto Analysis
Barcharts created by several procedures now allow labels to be oriented in a vertical direction, as in the plot below:

This is facilitated by a new "All" button on the Text pane of the Graphics Options dialog box that rotates all of the labels with a single click:

Contingency Tables and Crosstabulation
Two new tests have been added that are specifically designed to determine whether there is a significant trend across the levels of an ordinal factor: the Mann-Whitney U test (for 2 categories) and the Kruskal-Wallis H test (for more than 2 categories).

General Linear Models, Nonlinear Regression, Logistic Regression, Probit Analysis, Poisson Regression, Negative Binomial Regression, Zero-Inflated Poisson Regression, Life Data Regression,  Cox Proportional Hazards, Quantile Regression
Trellis plots have been added to each of these procedures to show how the response variable changes as a function of 2 or more variables simultaneously. A typical plot is shown below.

Poisson Regression, Negative Binomial Regression, Logistics Regression and Probit Analysis
Shaded confidence regions may now be added to the Plot of Fitted Model.

Factor Level Reordering
Factor level reordering using the mouse has been extended to the following procedures: Poisson Regression, Negative Binomial Regression, Zero-Inflated Poisson Regression, Logistic Regression, Probit Analysis, Quantile Regression, Ordinal Regression and Multinomial Regression.