Prior to starting any sort of observation or experimenting involving humans, you have to present your study prospect to the Institutional Review Board, or IRB. Your prospectus, if done well, should include any consent forms and permission forms that you must present to your participants. These consent forms must have:
- confidentiality agreement
- disclosure agreement
- anonymity agreement
- purpose and direction of the investigation
With this latter, you may want to explain whether your observational study is
- Cohort studies take place over time
- Descriptive vs. Analytic
- Prospective vs. Retrospective
If you submit the documents correctly, and as prescribed by the institution with whom you are investigating, the IRB should be no trouble and will more than likely guarantee approval. The process for IRB approval can take up to one month depending on the current institutional workloads for investigation.
Now, to your question on random selection. Random selection does not place any emphasis on specific traits about participants. However, it is a valid methodology and does reunite data. Yet, that data is more related to the external validity of the research. This means that it makes the investigation valid in terms of generalization. Think about the scientific method, the external validity of this method is that the procedure can be generalized and repeated regardless of whether the sampling was done randomly or specifically.
Internal validity is about the "inside job" of your experiment: did it give you answers that can help you make correlations? From your experiment, can you conclude that A causes B? Since some researches are experimenting with specific variables then the sampling might have to be more specific. For example, if you research the smoking habits of women who are shelter residents, you must extrapolate what specific behaviors you are looking for. If you want to start clean, then you will randomly select shelter women (after IRB) and then proceed to conduct your studies. If you conclude that shelter women tend to smoke more, your external validity is there, now, its time to dig deeper and find potential causes as to why. That is your internal validity. Hence, your project is valid enough in form for others to repeat, however, the causative factors as to why shelter women smoke more once they are sheltered would be what your internal validity will show. In external validity the two main steps is: draw a sample and generalize back. For internal the formula is A plus B causes C whether the sample is random or not. This is how researchers decide exactly what they want to study and with whom.
For further examples, refer to the Minnesota Center Against Violence and Abuse (MINCAVA) - I enclosed the webpage for your convenience. This organization conducts studies such as Housing and Battered Women: A Case Study of Domestic Violence Programs in Iowa. Also look into The Ford Foundation, which looks after women's rights and the safeguarding of battered women in shelters.