# Binary Logisitic regression to predict treatment type using time (year)?

#### Bcteagirl

##### New Member
Hello! I have some small data to work with, to look at whether one health centre has shifted from using treatment A to treatment B over time (Bivariate outcome). I have no covariates. I have been asked to conduct a very simple analysis to show that treatment A is decreasing over time while treatment B is increasing (They are non-overlapping). Looking at the crosstabs this is very clearly the case. (the relationship of time to the overall count of cases needing treatment is quadratic).

The data I have are counts by year (by treatment), 11 years of counts.

My question is how to attach a simple p-value to this relationship in SPSS.

I have searched through the internet and this forum. I am hoping it will be appropriate to use a binary logistic regression, using time as a predictor. It looks as though it is best to identify time as an ordinal variable (One variable for time, with levels 1-11). (In a previous analysis of number of cases over time there was little to no autocorrelation).

Thank you very much for any guidance you can give me.

#### GretaGarbo

##### Human
Why did you start a new thread? Wasn't it good enough in the old one? Besides, now you lost the connection to the data. (But people are very reluctant to open attached files.)

The data I have are counts by year (by treatment), 11 years of counts.
When I looked at the data (in the old post) there was 12 observations. Why 11?

The conventional way of dealing with counts is to use Poisson regression. But there is absolutely no information there to justify that it would have that distribution. (The data is certainly not binary so logit does not seems to be justified.)

Plot the data and look at it! The "quadratic assumption" does not seems to be
justified. (And it is not significant.) (And "time" is not only "ordinal", it is also an ratio scale variable.)

Besides, OP and others can read this depressing post ("Leif and Uri..").