# How can I predict a single binary outcome with multiple repeated measures?

I am using SPSS 25. I want to predict the probability of a binary outcome (DV): pass or fail a test at the end of “this” school year. My independent variables are

• Test status (pass/fail) at the end of “last” school year as a baseline measure
• Course grades or Grade Point Average (A=4, B=3...F=0) for “this” school year. This is a repeated measure where we have three observations prior to the outcome event
• Attendance data (number of days missed), which can be represented as a simple scaled variable (0-180) or as a percentage of days missed out of total days
• Demographic characteristics (e.g., age, ethnicity, gender)

I have already used logistic regression to establish the probability of passing given the baseline event (pass/fail “last” school year) and the first and/or second observation of course grades. The challenges I am still facing are how to include multiple IVs (such as attendance Or the third observation of course grades) when the IVs are highly correlated with each other. Right now I’m using SPSS and logistic regression, with pass/fail “this” year as the dependent variable, and then pass/fail “last” year in step 1. Then I add other IVs in subsequent steps as they occur (i.e., the first course grade in Step 2, the second course grade in Step 3).

On top of all this, my data are FLAT right now (e.g., all data for one student in one row). I might be able to build a stacked file for the analysis, but I cannot seem to think through how I deal with the baseline event And multiple IVs with different weights that happen at different times.

My ultimate goal is to be able to impose a formula onto my flat file that can be updated as new data get added or updated over time. Even if I have to do all the analyses using a repeated observations model and a stacked dataset, my users are viewing charts and reports based on the flat file - so that is where I need to have some sort of formula that can “feed” off data in the flat file as it gets updated.

Example: imagine the P of passing Your math test “this” year if you failed “last” year is .17. But if you begin “this” year and get an A in your math claSS, the probability changes from .17 to .42. On the other hand, if you failed the math test “last” year and you earn a C in your math class at the start of “this” year, the probability remains at .17. So I ultimately need an equation where new binary events (e.g., A in math = 1, else 0) can change the probability that is calculated in the flat file.

Man, I hope this makes sense to someone.

## 1 Answer

Consider to build the generalized linear mixed models (GLMMs) in SPSS. Since it was aforementioned that the data were "FLAT," before running the GLMM procedure, we have to use VARSTOCASES to convert the data to a long format. Here is an example you can refer to. https://www.ibm.com/support/knowledgecenter/SSLVMB_23.0.0/spss/base/syn_varstocases_examples.html

Analyze -> Mixed Models -> Generalized Linear... Select a variable to denote the Subjects, and "Course grades or Grade Point Average" to be the Repeated Measures. Then you can move on to the other settings. Make sure you toggle "Binary logistic regression." I think you may also want to go to Model Options/Save Fields, and check "Predicted values" and "Predicted probability for categorical targets."

Based on your Example, as far as I can understand, you may need to build different models by including different number of levels in the repeated measure, and predict the probability for each model.