5.4 Placebo Tests

Which Results Should We Believe? Role of Placebo Tests

Cross-section comparison

## were there pre-existing differences between the groups?
mean(basqueBefore$gdpcap) - mean(othersBefore$gdpcap)

## [1] 1.616077

Before-and-After design

## was there a change in a group we don't think should have changed?
mean(othersAfter$gdpcap) - mean(othersBefore$gdpcap)

## [1] 3.161306

What about the Difference-in-Differences design?

## here we go back in time even further to examine "pre-treatment" trends
## we want them to be similar
(basqueBefore$gdpcap[basqueBefore$year == 1972] -
 basqueBefore$gdpcap[basqueBefore$year == 1955]) -
    (mean(othersBefore$gdpcap[othersBefore$year == 1972]) -
     mean(othersBefore$gdpcap[othersBefore$year == 1955]))

## [1] 0.07147071

These “placebo” checks are closest to zero for diff-in-diff, so we may believe that the most.

Thanks to Will Lowe and QSS for providing the foundations for this example