10-16-2014 10:23 AM
Hello! I am analyzing data from a national survey. In order to account for sample allocation and survey design, all estimates are weighted using survey weights in order to reflect the age and sex distribution of the household population aged 15 or older in the ten Canadian provinces. Furthermore, variance estimations such as standard error and 95% confidence intervals are calculated using bootstrap weights provided with the data.
Given the nature of the data, we are using 'proc surveylogistic' in order to carry out a multivariate regression analysis. While the output provides the 95% CIs for the adjusted odds ratios (ORs), it does not provide the p-values. Note, I was told that the p-values for the parameter estimates that are provided in the output (i.e., Pr>Chi Sq) can be used for this purpose however, they are not in sync with what I would expect to be significant (or not significant) when compared against the 95% CI for the adjusted ORs. We've searched online and can't seem to find a specific procedure that will calculate the p-value for the adjuste ORs. Does anyone know how to do this? Thanks in advanced for your time!
10-16-2014 10:47 AM
A kludge method would be to get the log OR, and the log upper and lower bounds for the adjusted OR. From that, you could estimate the standard error, and from that and the estimate, you could calculate a Wald chi-square. But that is kludgy, and makes a lot of assumptions.
05-31-2017 01:05 PM - edited 05-31-2017 01:06 PM
Hi, I think you're having this problem because SAS is using effect coding for the coefficients but dummy variable coding for the Odds Ratios. By specifying " / param = ref" in the class statement you can make sure that SAS uses dummy variable coding for both the coefficients and ORs.
UCLA's IDRE has a nice explanation of this: https://stats.idre.ucla.edu/sas/faq/in-proc-logistic-why-arent-the-coefficients-consistent-with-the-....