Data validation What does data validation mean in a research context? For example I am doing a linkage study, linking developing congenital anomalies with taking antibiotics in pregnancy. How can I validate the results?
 A: Data validation can mean several things, all of which are important for a study like the one you have outlined.
First is whether the codes entered into clinical electronic records, for example, actually represent the true clinical data. These codes are often entered by people with minimal clinical understanding and are found to be erroneous on review. For that validation you need to have access to the original clinical records and the codes entered into the clinical record system and compare a sufficiently large sample to judge the problem.
Second is whether the records of the particular clinical institution you are examining is representative of a broader population of interest. Problems can come from distortions in data entry so that cases in the records are not even representative of the institution: for example, will you have records of antibiotics not prescribed by the institution but by an outside clinician? That validation can be harder, as you have to examine more closely how cases are entered into the records of the institution. Furthermore, the particular institution might have a unique patient population that is not seen elsewhere. Incorporating multiple institutions helps with that problem, but makes analysis of the coding-error and other within-institution problems even harder.
I'll omit issues of model validation based on the data you obtain, as that is well covered elsewhere on this site.  The cross-validaton tag you originally applied to this question has to do with validating the model you develop from the data, not with validating the data themselves.
The ultimate validation of your data and resulting model comes when you use the results of your analysis of the clinical data to design a prospective study, in a way that minimizes the problems noted above, including institutions other than those involved in your initial study, and then replicate your results.
