When chosing a study design, you should think about the following issues.

What is your hypothesis? What do you believe will be the outcome of your study?  The results you expect to see will greatly influence your study design.  You should set up your study to insure that, once the study is complete, your hypothesis has been either proven or disproven.

Will you do an intervention?  If you only want to describe the natural history of a population or a disease process, you will be doing an observational study.  If you make an intervention, you will be doing an interventional study.

Can you have a control arm to your study?  A control arm is very important to help prove or disprove your hypothesis.  This the the group of subjects that does not recieve your intervention in an interventional study.  In an observational study, a control group may be a naturally occuring population that differs from your study population.

Will you use pre-existing data or generate new data?  Perhpas data already exist that might answer your study question.  If so, using this data might be cheaper and faster than collecting new data.  Cohort studies often use pre-existing data and can be either retrospective or prospective in nature.  Interventional studies will always generate new data.

Will you look at subjects over time, or at a single point in time?  Cohort studies and interventional studies look at subjects over a given peroid of time.  Cross sectional studies collect data on exposures and outcome at a single point in time.

Will you have a control group? A control group is made up of subjects that do not have the exposure of interest.  The ideal control group differs from the experimental group by ONLY the exposure of interest (a toxin, a drug, an educational intervention).  If you have a good control group, any difference in the outcomes between the control and experimental groups is likely due the the exposure of interest.

Can you randomize subjects? In an interventional study with multiple groups, how a subject is assigned to a group is very important.  By randomizing which group a subject enters, selection bias is minimzed and the study results will be more robust.  Randomization can be done in a number of ways.  Assigning every other subject to a group is a common method.  Assigning patients based on the last digit of their telephone number or based on the day of the week can also be done.  Random number generators are probably the least biased method and are available free online. 

Can you blind subjects, the investigators and the statistician? Blinding helps to reduce bias in a study.  When the subject is blinded, it helps reduce any effect that would result simply from knowing you are being studied (The Hawthorne Effect).  Subject blinding is usually done by using a placebo.  The placebo used will depend on the type of intervention being studied.  Blinding the investigators prevents any subjective influence from creeping into the data collection.  When both the subjects and the investigators are blinded, the study is referred to as doubled blinded.  Blinding the statistician insures that the data analysis is done in an unbiased manner and without looking for a given result.