Estimating Causal Effect in Absence of Pre-Treatment Data

[saqib hussain]

Hi everyone,

I am analyzing the impact of a policy that was initially implemented in 18 districts in 2013 and later extended to an additional 18 districts in 2017. My dataset is a balanced panel covering the years 2014 to 2022. The policy was phased in due to budget constraints, as noted in the policy documents.

1. Given this phased implementation, should I treat the rollout as a random implementation of the program?
2. What methodology would be appropriate for my analysis in the absence of pre-treatment data? I would appreciate if someone can explain how can I use the district and year fixed effects.

Thank you in advance for your assistance.

[Jared Greathouse]

Depends. I lean towards Difference-in-Differences. I would get a longer time series though

[Rich Goldstein]

esp if you have some pre-2013 data, you might be able to analyze this as though it was stepped wedge trial (but you don't really give us much info); see, e.g., Twisk, JW, et al. (2013) Different methods to analyze stepped wedge trial designs revealed different aspects of intervention effects," Journal of Clinical epidemiology, 72: 75-83; or, Twisk, JWR (2021), Analysis of data from randomized clinical trials, Springer (and, yes, I realize you don't have an RCT but that speaks more to how you interpret and write up the results)

[saqib hussain]

Thank you for the responses!

I am analyzing the impact of Land Record Digitization on corruption in land departments. The Anti-Corruption Establishment has data available on its portal from 2014 onwards. The land record digitization occurred in 2013, initially in 18 districts, and was expanded to all 36 districts in 2017.

I am seeking advice on how to define the treatment variable for my analysis. Specifically, I need guidance on how to appropriately account for the phased introduction of digitization across districts and over time. The outcome variable in my analysis is the count of corruption incidents reported in land departments.