"A simplified time-series analysis for evaluating treatment interventions": A rejoinder to Blumberg.

Dodd (1984) wrote a short reply to another scientist. The paper did not collect new data. It showed, with math, that the C statistic can fool you.

The C statistic counts how many data points fall above or below a line. If the line is perfectly straight, the count says nothing about change.

The C statistic equals zero when data line up in a straight slope. Zero looks like "no effect," even if the behavior is clearly improving.

In other words, the tool can say "nothing happened" when something really did.

Mulvaney et al. (1974) and Michael (1974) said the same thing ten years earlier: most summary numbers break in single-case work. Dodd (1984) gave a fresh example.

Campbell (2004) later tested four newer effect sizes. All four still disagreed, proving the problem is alive.

Barnard-Brak et al. (2020) offered a fix: use Bayesian N-of-1 models instead of simple counts. Their method sidesteps the straight-line trap that W warned about.

If you plug numbers into a formula without looking at the graph, you can miss real change. Always pair visual analysis with any summary number. If the data look like a perfect ramp, skip the C statistic and pick a tool that measures slope or level, not just overlap.