Time series or just repeated measures?

#1
Hi All,

I have data on 7 clusters (ie, sites). The date are monthly counts (there is no zero inflation in most of the outcomes of interest...most clusters have a count in the 10s-100s for each month) , from Jan 2004 to May 2010, however only one cluster has data before March 2006. By August/Sept 2006 all clusters are reporting, though there is some missing data (essentially missing at random).

I should say that most of the outcomes of interest are monthly counts. One or two of them are actually cumulative, so a scatter plot shows a fairly monotonic rise over the months.

Here's the deal. At one of the timepoints, say June 2007, there was a significant change in the way all the clusters did business, as a result of a change in policy, effective across all clusters.

I am wondering about ideal ways to test for changes in the outcomes as a result of this policy change at this timepoint.

I am not at all trained in time series but I am trained in linear and generalized linear modeling. I am not a statistical newbie, but even with a masters in stats and 10 years of experience I am still not 100% sure which route I should take.

Or rather, I should say that my training and experience tell me to do a repeated measures on the clusters, use an autoregressive correlation matrix, and look for some structural slopes.

But, you know what they say about toolboxes? If you've got a tool box of various sized hammers and screwdrivers then every problem looks like various nails and screws.

So, rather than fall into that trap I thought I'd ask the statistical community for suggestions. Also, I realize there are likely many ways to skin this cat (gross) so I'm open to examining all options.

Thanks, and please let me know what else I can tell you.

Sabby