---

Experimentation in the social sciences (psychology, sociology, education, ... etc. ) is properly referred to as "quasi-experimentation". It is quasi-experimentation because anytime you deal with human beings, you can not isolate all extraneous variables from the variables you are actually studying... With physics experiments, you really can isolate "other causes", but with human beings, if you structure an experiment (unless it is a one shot deal, of a very simple nature) people go home, do other things that may affect the experiment. If its an experiment with more than one stage, some of the participants may not show up at the later stages........ this is called "mortality". And it may be that the people who show up are different than the people who don't in some important way connected to the subject of the experiment. Even if you have a simple one-shot experiment, exery human being is completel different in countless ways, it may be hard to generalize and infer causation....

However inferring causation is a major goal of experiments and if you want to infer causation, it is hard to do it with another methodology....

Another issue with quasi-experimentation is that you need the lawful consent of the people you are experimenting on....

A quasi-experiment is exactly what it sounds like. It is an experiment involving human beings possibly in their usual sociological context. You investigate the effect of some variable by conducting an experiment where you expose participants to that variable (that they presumably wouldn't be exposed to ordinarily). Most of the usual ideas about experiments still apply... treatment, control group, pre and post-tests.... etc... except that because it involves human beings there are an enormous of things you can't really control, like what they do in their spare time, that they may drop out of the experiment partway through.... that they may talk with each other during or between sessions... All these problems that are associated with the complexities of human beings and their social complexities are what make it a quasi-experiment and not an experiment.. You can avoid these complications entirely, but you should try to account for them.

Here is an example suggested by one student in the class:

"If a class is randomly divided and taught in exactly the same manner in every respect except access to technology. Then the researcher can try to put together the causality of any differences in behavior exhibited. Examples might include test scores, facility with new computer programs, likelihood to use a computer to type a paper, edit work, research a subject, etc. Although the researcher cannot eliminate all the complexities of life as a physical scientist can, in a quasi-experimental setting, the researcher can minimize as many factors as possible and try to make the hypothesis as straight forward as possible."

This is how I reacted to this:

This is a good example and deserves to be looked at a little more closely. I think I'll copy it and put in the mini-lecture on quasi-experimentation. ... not only minimize the extraneous factors but also try to account for the ones that can't be minimized...
> also there is a question of ethics..... will half of the class be receiving less good an education.. Can they get the same chance in the next unit maybe?

Let's talk about some quasi-experimental designs.... All of the designs have some problems. That doesn't mean they shouldn't be done... it just means if you do them... you have to understand and discuss these problems..
Of course the less great the problems, the more you can infer from the experiment.

I am going to use some coded symbols here...

'X' means treatment. 'O' means observation. We will use the convention that chronological time goes from left to right.... time------

The one-group post-test only Design: (One group receives treatment, then later is post-tested.)

______
X O
---------

You treat some group and then observe changes afterwards. Problems: Lack of pretest observations , one cannot easily infer that the treatment is related to any kind of change.. also lack of control group.

The posttest-design with nonequivalent groups: (Two groups, one gets the treatment, the other doesn't then both get a posttest)

_____
X O
-------
O
--------

Problem: Because of the lack of pretests, the post-test differences between the groups can be attributed to either to the treatment or to the selection differences between the two groups.

The one-group pretest-posttest design: (One group gets pretest, then gets treatment, then gets posttest.)
__________
O1 X O2
__________

The problem here is illustrated with an example. Suppose you are doing a study where the supervisory style of a foreman is changed in a work setting. The productivity goes up after the change. Does the increase in productivity have anything to do with the change in supervisory style? Maybe other things happened during the same period. What if a salary increase occured during the same period? Maybe the change is the result of that.

One more...

The untreated control group design with pretest and posttest: (Two groups, both gets pretest, then only one group gets treatment, then both get posttest)

_________
O1 X O2
----------------
O1 O2
___________

This last one has more problems than the last.... but it still has problems.... Suppose the two groups are by chance unequal, one group is smarter. Then improvement may because of learning that had nothing to do with the treatment. That smartness may not be detected by the pretest... for any number of reasons.