I have a dataset of the number of drug related events (e.g. arrest, non-fatal overdoses) a population of individuals have had. My aim is to see if a certain policy intervention that occurred on 1 June 2016 has had a significant impact on the number of drug related events that these individuals (approx. 4,000 people) have had. I thought of using interrupted time series analysis to see if there is a reduction in drug events for these individuals from the 2 years prior to the intervention compared to the two years after.

However, most (about 60%) of the individuals in this population have only had one drug event in the 2 years prior to the intervention. In fact, I have their data going back much longer than this for some – up to 10 years – and it seems these are rare/one off events for these people.
Using interrupted time series analysis on the 40% of individuals who have frequent events seems to make sense – they have the potential to show a decline e.g from a rate of 6 events to 4 events post intervention. However, for the 60% who have only one event, they don’t have much of an opportunity to show a decline post intervention. Even if these individuals did decline from 1 to 0 events in the 2 years post intervention, it seems like I would need to strongly qualify this finding with the fact they may have only had 1 event in their lifetime anyway, so no events over the following 2 years is hardly surprising.

Are my thoughts right about this? Should I not be using Interrupted TS? If not, is there another technique you could recommend to analyse the data?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.