a question regarding single group interrupted time series

[SY Kim]

I'm using itsa command and have a question regarding single group interrupted time series.

I have six time points.

In the figure below, it seems that the slope has changed after the intervention.

However, what the data indicates is that it's not significantly different. (beta coefficient for _x_t2020)

Please give me any advice for interpreting the results.

Thank you.

[Jared Greathouse]

My advice is to not use ITSA with 6 observations. You're almost literally better off doing just a t test. Better to use a simple method that you can try to defend instead of a more complex method where its main assumption isn't met (in this case, a long pre-intervention time period).


In other words: you're asking people to believe that three data points are a good proxy for the post-intervention counterfactual, and I'm so sorry, that's just not something anyone will buy, even in an undergrad level causal methods course. Perhaps I'm being unreasonable, and maybe Ariel Linden feels different from me, but my sincere advice here is to keep away from ITSA if its basic identifying assumptions aren't met. Specifically,

Sequential measures of the outcome should be available both before and after the intervention. There are no fixed limits regarding the number of data points, as the power depends on various other factors including distribution of data points before and after the intervention, variability within the data, strength of effect, and the presence of confounding effects such as seasonality... Given the requirement for a relatively long time series, routine data are often most appropriate in ITS studies... The example dataset has 59 months of routine hospital admissions data with 600-1100 ACEs at each time point.

So in other words, even if the estimator runs mathematically, any interpretation we could try to offer just won't be meaningful here, in my opinion.