Linear regression is a great tool when your outcome variable is test scores or loan amounts or another continuous variable. But sometimes, your output is a Yes or a No. That type of outcome is known as dichotomous.

You still can do something similar to linear regression because some super smart stats dude awhile back came up with a way to mimic linear regression with a dichotomous outcome variable.

To do logistic regression SPSS, you need to have the “regression models” add-on program. You also have to understand your data and do a little prep work on it.

As with any analysis, your data must fit some basic assumptions:

1. You don’t have a lot of missing values. If you do, you need to create dummy variables so you can test the “missing” category.

2. You have enough records. (Some suggest a minimum of 50.)

3. You don’t have strong collinearity between your independent variables. If you do, you might have to create some composite variables. You can check this by running a correlation matrix on all your independent variables.

Prep your data:

1. All of your categorical variables must be made into dichotomous variables. For example, if you have four races, you’ll need a category for white (1,0), black (1,0), etc. But you won't include all categories. You leave one out as your “reference” category against others will be compared.

2. Don’t leave anything out. Include every variable you have at the beginning. You may not use it all in the end, but you must make sure nothing else affects your model.

3. Beware of data within data. If, for example, you’re looking at traffic stops for multiple police departments, your model may be better explaining department differences than differences in individual stops. To account for this, create a dummy variable for every department to eliminate any affect by department.

4. Logistic regression is a better testing tool than reporting tool. Once you figure out your key variables in your model, go back and run crosstabs. Use those figures for graphics and reporting; they will make more sense to your readers.

Once your data is ready, you can do the logistic regression. Here’s how:

Go to ANALYZE | REGRESSION | BINARY LOGISTIC

SPSS will prompt you for the DEPENDENT and INDEPENDENT (OR COVARIATE) variables:

SAVE: If you check PROBABILITIES under SAVE. SPSS will save the probability that each variable will have the outcome.

OPTIONS: Check the Hosmer and Lemeshow Test for goodness of fit.

What do those tables mean?

You get lots of output with a logistic regression, what's most important is the table called VARIABLES IN THE EQUATION. The column labeled "Exp(B)" is important to your analysis. That basically tells you how much more likely a variable is to have a particular outcome than other variable. The other important column is "Sig." that tells you if the information means anything. If your Sig. - or significance - is less than .05, that means your data is meaningful. In other words, it couldn't just occur by chance.

What do the diagnostic tests mean?

1. Pseudo R Squared. Because data in a logistic regression does not form a line, there is not real measurement like R squared in linear regression. However, several researchers have come up with what are called "pseudo" R squared calculations that approximate how much of the variation in your dependent variable is explained by your model. Keep in mind that these are rarely as high as they are in a typical linear regression.

2. Hosmer and Lemeshow Test for goodness of fit. This is a complicated statistical measure that tells you how good your model is. What you need to look at is the significance measure. If this measure is less than .05, your model does not fit your data very well. You want this to be larger than .05.

Stay tuned. I'll add an annotated output soon!