From the course: SAS Essential Training: 2 Regression Analysis for Healthcare Research

Basic PROC LOGISTIC output - SAS Tutorial

From the course: SAS Essential Training: 2 Regression Analysis for Healthcare Research

Start my 1-month free trial

Basic PROC LOGISTIC output

“

- [Woman] Here we are back in SAS. I still have the exercise file from the last movie open, which is called 600_Logistic Models 1 and 2. In the last movie, we just ran some prog logistic code. We'll start this movie by looking through the output. Here is the top of the output. You probably won't really look at this much, but it's good to use if you are using different data sets because the number of rows is always on the top. That way, you can make sure you are using the right data set in the analysis. See this note? That's a good note. It means you can rest assured that the odds ratios that come out of this model are positive if they are associated with reporting asthma and negative if they are associated with reporting no asthma. See this table here that says Model Fit Statistics? We'll come back to this one later when we run some models and need to consider model fit. I know what you are thinking. Where are the parameter estimates? Where are the slopes? Prog logistic output is a little wordy at the top, but here we go down from here under Analysis of Maximum Likelihood Estimates. That's the parameter table. And remember, these slopes are on the log odd scale. See this slope? It says 0.3729, but it's on the log odd scale. We need to turn it into an odds ratio and then put confidence intervals around that odds ratio. How do we do that? Look, SAS to the rescue. That's what's so cool about prog logistic. See how it automatically produces the odds ratio estimates and their 95 percent confidence intervals? Here's a question for you to consider. In this model, is DIABFLAG statistically significant? Well let's think about it. Is one contained within this odds ratio? That's what you have to ask yourself. We see that the answer is no. The lower limit and the upper limit and of course the estimate in between them, all these are above one. So we can tell from the confidence interval that this odds ratio is statistically significant. But you know what's wrong with it? It's not adjusted for any confounders. So let's go back to our code and do some more work. See this second prog logistic? In this one, I added male and all the age indicator variables. Let's highlight and run this code and see what the output looks like. As you can see, just adding a few variables really makes the output get long. See these parameter estimates? Let's look at the P values on these slopes. See here, only DIABFLAG and male are statistically significant in this model. And so, when we look down here, the only confidence intervals that do not contain one are the ones for DIABFLAG and for male. Given that the outcome is asthma, can you tell if DIABFLAG and male are associated with increased risk of having asthma or decreased risk of having asthma? If you said increased risk for DIABFLAG and decreased risk for male, you were right. That is because for DIABFLAG, the entire confidence interval is above one and for male, the entire confidence interval is below one. Great, now that you are fluent in prog logistic, let's move on to planning our models.

Contents