5 Wide to long and back, and handling spell data

Wide format and long format data, where we have alternately many observations per individual in one record (wide) and many records per individual each with one observation (long) are both useful ways of representing repeated data. This applies both for panel data (one observation per wave) and continuous-observation spell data (each spell is an `observation' or case even if it is retrospectively reported).

Each type has advantages: wide is good for looking, for instance, at specific pairs of observations (cross-tab of wave 1 versus wave 2, logistic regression on attrition by wave 2 using wave 1 data, etc. ). Long format makes it easier to look at wave-n/wave-n+1 transitions, among other things.