The original presentation

Analysis of Psychosocial Treatments for Cocaine Dependence

The document

The analysis was originally presented as a slide deck covering the trial background, the four arms, the ANCOVA models at both timepoints, the interaction analyses and the longitudinal model.

Presentation (PDF) Slide deck Executed SAS output

How the site relates to it

The analysis page is not a restatement of the deck. Every model in it was recomputed from the trial data, and the published figures reproduce exactly: R² of 0.343 and 0.252, F of 37.45 and 27.34, treatment p of .0110 and .2580, and the adjusted subgroup means to two decimal places.

What the site adds is the material a reader needs to weigh those numbers:

What the site adds to the original presentation.
AdditionWhy it matters here
Retention by armOnly 60% and 68% of randomized patients reach the two analyses. The comparison between timepoints depends on who remained.
Confidence intervals on every contrastA p-value alone does not show how large the plausible range of effects is at this sample size.
Assumption checks on the reported modelThe original tested residuals from a model without the covariate. On the reported model the month 6 violation does not appear.
Cell sizes with every subgroup estimateThe largest subgroup gap rests on 14 and 10 patients out of 16 cells.
An explicit statement about multiplicityThe subgroup finding is the most extreme cell in a large family of tests.
The derivation of the recommendationSection 6 of the analysis traces the chain from a null overall result to a subgroup recommendation, and shows all four arms for that subgroup rather than the two that generated it.

Where the site's reading differs from the deck's, the difference is one of confidence rather than of arithmetic. The numbers agree. The site is more cautious about what the subgroup results can carry, and more explicit about what the missing patients mean.

About the data

The analysis uses patient-level data from the NIDA Collaborative Cocaine Treatment Study, which is restricted and not redistributable. It is not in this repository and is not published here in any form. Everything on this site is aggregate.