You need to quantify the causal effect of a new pricing strategy launched exclusively in City A on January 1 2024. Daily customer-churn rates are available from January 1 2023 through December 31 2024 for City A (treatment) and five comparable cities that retained the old pricing (controls). Exploratory plots show strong weekly seasonality and autocorrelation that are common across all cities. Management requests an estimate that (1) removes macro trends shared by every city, (2) avoids relying on forecasts of City A's post-treatment values, and (3) yields valid standard errors even when observations inside a city are serially correlated. Which temporal-modeling framework best satisfies these requirements?
Estimate a two-way fixed-effects difference-in-differences model with city and date indicators and cluster standard errors at the city level.
Train separate AR(1) models for each city and infer the treatment effect by comparing the autoregressive coefficients of City A with those of the control cities.
Fit a seasonal ARIMA model to City A's pre-treatment data, forecast the post-treatment period, and treat the forecast error as the effect.
Apply a parametric Weibull survival analysis to individual customer tenure in each city and compare survival curves between treatment and control locations.
A two-way fixed-effects difference-in-differences (DiD) model compares pre- versus post-treatment changes in City A with contemporaneous changes in the untreated cities, automatically controlling for common shocks and seasonality through date (time) fixed effects and for city-specific level differences through unit fixed effects. Clustering standard errors at the city level addresses serial correlation within each city, giving valid inference even with daily data.
Building a seasonal ARIMA forecast for City A (Choice A) uses no control group, so any macro shock coinciding with the price change would be mis-attributed to the treatment. A parametric Weibull survival model (Choice B) targets individual time-to-churn and does not inherently difference out shared trends across cities. Separate AR(1) fits with coefficient comparisons (Choice D) capture short-run autocorrelation but provide no principled counterfactual or inference on the treatment effect. Therefore the DiD framework with clustered standard errors is the most appropriate choice.
Ask Bash
Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.
What is a two-way fixed-effects difference-in-differences (DiD) model?
Open an interactive chat with Bash
What does clustering standard errors at the city level mean?
Open an interactive chat with Bash
Why wouldn't a seasonal ARIMA model be appropriate in this scenario?