# Binary or Multinomial Logistic Regression in SPSS: Interpretation and Reference Categories

I am trying to analyze my data using Multinomial Logistic Regression whereby my dependent variable is a clinical outcome (sick vs healthy) and 1 independent variables (Factors) are in several categories.

The problem I have is trying to figure out how I can set one of the category as a reference group in SPSS.

I have done the inverse analysis by switching dependent and factor variables so I can set the reference but after now that I think about it, I realize it doesn't make much sense. I also compared the OR values from MLR to that of a 2x2 analysis but it is vastly different.

I also tried Binary Logistic Regression and created dummy variables for each category but I didn't have sensible values either.

EDIT: SPSS command and output [I have set the first category (1) in my independent variable as reference]

NOMREG Group (BASE=FIRST ORDER=ASCENDING) BY Status
/CRITERIA CIN(95) DELTA(0) MXITER(100) MXSTEP(5) CHKSEP(20) LCONVERGE(0)
PCONVERGE(0.000001) SINGULAR(0.00000001)
/MODEL
/STEPWISE=PIN(.05) POUT(0.1) MINEFFECT(0) RULE(SINGLE) ENTRYMETHOD(LR)
REMOVALMETHOD(LR)
/INTERCEPT=EXCLUDE
/PRINT=PARAMETER SUMMARY LRT CPS STEP MFI. • If your dependent variable is binary, binary logistic regression is the way to go. What do you mean by "1 independent variables (Factors)?" ... What variable presents a problem with setting the reference category? ... What specifically "doesn't make much sense"? Also, how about showing your commands and your output so volunteers here can help you interpret and troubleshoot. Jul 14, 2014 at 21:43
• Thanks, I will edit my original text to include some output and commands!. Just to be clearer, my independent variable is categorical (7 distinct groups). I would like to use 1 of the group as a reference category. Jul 14, 2014 at 22:26

You can achieve what you are looking to do via the following.

1. Use binary logistic regression.

2. Assign your binary Status (sick vs. healthy) variable as the dependent. Recode if necessary so that sick = 1 or healthy = 1 (and the other is 0), depending on whether you are more interested in modeling the log-odds of being sick or of being healthy.

3. Assign a reference category to the Group variable using the Contrast command. Help files or a syntax guide will aid you in choosing from among options such as Indicator or Deviation contrasts (Indicator will probably be most convenient) and in the mechanics of assigning one category such as GCA as the reference to which others will be compared.

4. Creating dummy variables to represent a predictor such as Group is useful in some instances but is probably not necessary here. SPSS will create these dummies for you as part of the contrast you specify. Later, if you need to use regression output to create a predictive equation, there is a shortcut to doing so without creating dummies which I can share with you separately if need be.

EDIT - to assign a specific group as the reference category:

frequencies group


This will show you the order of categories as SPSS "sees" them. Let's assume GTG is third. Then GTG can be assigned as the reference category by using this subcommand in the regression:

/contrast (group) = indicator (3)


Now, assuming "healthy" is coded as "1" for the status variable, each "group" coefficient in the regression, when exponentiated, will tell you the ratio between that group's odds of having a healthy outcome and the GTG group's odds of having a healthy outcome.

• Thanks for the clear instructions. I will try this. Just to be sure, if I am interested in modeling odds ratios for sick, I should code the sick samples =1? Also, when I use the Contrast command, it seems that I can only choose the first or last as a reference category in the Indicator contrast. If for example, I need to set the reference category to the 3rd group, does this mean I would have to recode everything with the reference group=1 ? Thanks for your help again, I am fairly new to Statistics in general. Jul 15, 2014 at 17:33