Relative influences in time series data – sth better than distributed lags analysis?

[Chris Preager]

I am trying to ascertain whether civil society drives state reform, or whether civil society responds to state reform in repressive contexts*. The Polity IV data has excellent indicators of each over time.

To test which effect is greater, I tried distributed lags analysis, testing the correlation between the two variables at different time lags. However I only want to analyse shifts in one direction (towards democracy), for countries that are initially repressive, (so asking whether positive transitions in one are proceeded by the other). Is there a better method?

Any guidance very gratefully received,
Chris

* In my interviews and surveys, the international community consensus is
that the state must first open ‘associational space’ before civil society can be effective, whereas local peacebuilders describe civil society forcing reform on a reluctant governments.

[Chris Preager]

My crude idea at the moment is to create a column that is the difference between current and next period for each variable, filter out any negative entries in the independent variable, then test the correlation between these at different lags.

I assume there must be a more sophisticated way of doing this, as for example filtering the independent variable makes the analysis lop-sided when I attempt to measure the converse influence. Also, I’m not sure how to interpret the results – perhaps the probability that a positive transition in one is followed by a positive transition in the other, (regardless of magnitude), would be a clearer statistic. However I can’t find a suitable method or discussion of considerations.

I got the highest grade at A-level maths (college level) so can do my own learning given a pointer. Conversely if you think I’m being over-ambitious please do say so.

[Author]

Posts