Controlling for Clustered data and Autocorrelation in regression

[mccormsa]

Hello! I am trying to run an interrupted time series analysis on admissions data from four clinics. The dependent variable is the total count of all visits (aggregated across the four clinics) and the independent variables are week, interruption, and the interaction variable. The dependent variable appears to be normally distributed, so I have been using proc autoreg in order to account for autocorrelation since it is time series data. However, my advisor suggested that I also need to account for the clustering of the data (since it comes from four separate clinics). I am not sure how or what procedure is available to account for both autocorrelation and clustering? Any advice would be greatly appreciated.

[sbxkoenk · Accepted solution]

Hello,

I have moved your question to the 'SAS Forecasting and Econometrics' board.

I would try to use PROC PANEL (SAS/ETS) or PROC CPANEL (SAS Econometrics in SAS VIYA).

Use one of these options in PROC PANEL :

Alternative Variances Options
CLUSTER	Corrects covariance for intracluster correlation
HAC <(options) >	Specifies a heteroscedasticity- and autocorrelation-consistent (HAC) covariance
HCCME=	Specifies a heteroscedasticity-corrected covariance matrix estimator (HCCME)

If you want to stick to PROC AUTOREG , PROC AUTOREG has something similar.
See this option :

COVEST=OP | HESSIAN | QML | HC0 | HC1 | HC2 | HC3 | HC4 | HAC <(…)> | NEWEYWEST <(…)>

on the MODEL statement !

Good luck,

Koen