"An analysis-of-variance model for intrasubject replicaiton design": some additional comments.

Mulvaney et al. (1974) wrote a short, sharp note about numbers.

They looked at how people were using ANOVA on single-subject reversal designs.

The authors said that ANOVA misses the point because the data points are linked over time.

The paper says ANOVA treats each score as independent.

In real ABA graphs, yesterday’s score shapes today’s score.

They urge us to swap ANOVA for time-series tests that handle this serial correlation.

Christophersen et al. (1972) had pushed the opposite view: ANOVA is fine for reversal data.

Mulvaney et al. (1974) and Lydersen et al. (1974) answer together—both reject that stance, forming a direct-replication pair that warns the same year.

DeHart et al. (2019) extends the warning by showing mixed-effects models give the same benefit without old ANOVA baggage.

Dodd (1984) later offered the simple C-statistic as a time-series tool, proving the idea stayed alive.

If you graph one client’s behavior across phases, do not run a one-way ANOVA in Excel.

Use visual analysis first; if you need stats, pick time-series or mixed-effects options that respect the order of sessions.

This keeps your conclusions honest and your graphs clean.