Erratum: A comprehensive set of postestimation measures to enrich interrupted time-series analysis

The article “A comprehensive set of postestimation measures to enrich interrupted time-series analysis”, by Ariel Linden (Stata Journal 17: 73–88), contains the following errors.

The formula for computing the treatment group’s difference in trends between the preintervention period and the second-intervention period is incorrectly written as β ₃ + β ₅ +β ₇ +β ₉ +β ₁₁, which appears in the eleventh sentence of paragraph 2 on page 77 and in table 1 on page 81. The correct formula is β ₃ + β ₇ + β ₉ + β ₁₁, and the corresponding generic lincom specification for this estimate is

where trp = treatment period.

The formula for computing the difference between the treatment and control group in the differences in trends of the preintervention period and the second-intervention period is incorrectly written as β ₅ + β ₇ + β ₁₁, which appears in the last sentence of paragraph 2 on page 77 and in table 1 on page 81. The correct formula is β ₇ + β ₁₁, and the corresponding generic lincom specification for this estimate is

where trp = treatment period.

The changes in these two formulas affect the results of the example, which are presented in table 2 on page 85. The correct lincom specification for the treatment group’s difference in trends between the preintervention period and the second-intervention period is as follows:

The correct lincom specification for computing the difference between the treatment and control group in the differences in trends of the preintervention period and the second-intervention period is as follows:

To clarify how these two estimates are derived, I reproduce figure 1 from Linden (2017) and add text to show the lincom specification for each slope (trend) estimate (for treatment and control at each time period).

Figure 1.

Multiple-group interrupted time-series analysis with Newey–West standard errors and two intervention periods; added text presents the lincom specifications for producing slope estimates for each period and study group.

The treatment group’s difference in trends between the second-intervention period and the preintervention period is

However, _b[_t] + _b[_z_t] cancels out in the formula, leaving the more succinct specification presented above.

The difference between the treatment and control group in the differences in trends of the second-intervention period and the preintervention period is

which shortens to _b[_z_x_t1982] + _b[_z_x_t1989] after cancellations in the formula.