Best practice when merging waves from non-panel survey This is more of a data management question.
When merging several waves of a survey where the same questions were asked in multiple years -- but each time of a new sample -- should the data from a question be merged into the same variable or separate variables?
Since each wave was administered to a new sample, the unique (person) identifier would be new for each wave and there is no risk of analysing answers from multiple waves as though it were a panel dataset. But might using a single variable (say q1 rather than separate w1_q1 w2_q1 w3_q1) be confusing or helpful?
 A: This is a distinction between what is sometimes called wide format and long format for panel data. Most frequently in wide format the data matrix is represented as;
Unit_ID Q1_Wave1 Q1_Wave2 etc..
1       1   2
2       2   1

Whereas in long format one typically represents the data as;
Unit_ID Wave#   Q1
1       1   1
1       2   2
2       1   2
2       2   1

In the end, any reasonable data management system (either statistical package or database) will allow transformation between the two in relatively few steps. 
In the case in which you have the same units for all waves wide format is defensible (although each has advantages in particular applications). In your particular case though, you don't have the same units for all waves. So in the end your wide format database would look like this (where I use . to represent missing data);
Unit_ID Q1_Wave1 Q1_Wave2 etc..
1       1        .
2       .        1

This is problematic, as you will likely want to make comparisons to the responses between waves, but because the data is not matched most programs will drop cases listwise for many different types of tables or comparisons. That is, if a case is missing any values in a row, it will be dropped. By definition of the nature of your survey, every row will have missing values.
Hence I would suggest the long format is more appropriate, and to make tables or comparisons one just has to use a statement in whatever program that makes comparisons by waves. So I would suggest you utilize the format;
Unit_ID Wave#   Q1
1       1   1
2       2   1

Which does not artificially insert missing data by definition. Although a minor distinction, it is hard to think of situations where wide is a better choice in the type of panel design with different units in each wave (but going between each isn't typically that arduous a task, so if need be you can transform from long to wide).
