Excellent advice, Thomas. Reasons for censoring (dropout and others) have to be explicitly explained to justify inclusion or exclusion. If the dropouts are some of the girls for reasons pre-specified, it is less problematic.
------Original Message------
Ajit gave good advice. I would like to make explicit, as I shall discuss further, that statistical analysis may have to account for the schools that dropped out.
The schools that dropped out present one issue and their replacement another. The former issue is how school dropout and girl dropout, the outcome, may be associated with each other in the population of schools. Unless there is good reason to judge any such association to be negligible, statistical analysis must allow for it. And, since non-negligible association may have affected assignment of replacement schools to intervention, it might not be practicable to include them in the analysis.
The latter issue is how knowledge of the data may have affected school replacement. The consequences are mostly parallel to those of association between school dropout and the outcome. The main difference is that such an effect on school replacement would not affect analysis based on only the 48 schools originally selected, without the replacement schools. But, unless there is good reason to judge effect on school replacement to be negligible, statistical analysis based on all 50 selected schools would have to allow for such an effect. And it might not be practicable to do so.
Finally, if it is unknown whether both association between school dropout and outcome and dependence of school replacement on knowledge of the data are negligible, secondary analysis under an assumption of negligibility might be appended.
------------------------------
Thomas M. Davis
------------------------------