Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
Drexel University or The Math Forum.


loom91
Posts:
17
Registered:
4/25/06


Re: Help with a school project, a statistical survey on education
Posted:
Jul 23, 2007 8:58 AM


On Jul 23, 8:06 am, Bob <bbx107....@excite.XYZ.com> wrote: > > Molu, > > Interesting project. You are so right to note that a realworld > project is much more complex than doing some calculations. > > An important point is what I might called "impeccable honesty". Report > what you do, and report it honestly, without exaggeration. There are > limits to what you can do, for various reasons (obviously including > time constraints). To the extent you accurately describe what you do > and its limitations, then you have made a contribution. It does not > matter whether you do or don't find some statistically significant > effect. Even if the stats suggest that you do, the conclusion may be > undermined by hidden assumptions; if they are hidden assumptions that > should be fairly obvious concerns, then it really is your obligation > to point them out. > > As a small example from what you wrote above... You said that a > certain comparison of scores would allow you to say "... then we may > say that > girls prefer humanities more over other subjects than boys.". No, not > at all. It is about "scores" or "performance", not about "preference". > You have given an example of how you tried to eliminate one > "confounding" variable. But, for all you know some of what you see > will be due to one teacher who is different. It is almost impossible > for you to deal with that. But it is a reminder of the issue of hidden > variables. > > What you are doing is science. A well done statistical survey is a > good scientific project. One asks a question, tries to formulate an > approach, and then collects some data. In the real world, it is common > that the initial effort leads to more questions than answers; it is an > iterative process. You won't have time for that, but you can remember > that how you present the work and how you analyze the results you > obtain are as important (more important!) than whether you find any > particular result. So emphasize good clear thinking at all steps. > Ending with a list of questions that come up is probably a sign of > success. > > There are some web sites that offer calculation tools online. They > also offer various degrees of explanation. I have no direct experience > with them, and thus no advice about which are good or easy to use, or > whatever. So you might just explore. Here are some I am aware of: > > http://faculty.vassar.edu/lowry/VassarStats.htmlhttp://www.ruf.rice.edu/~lane/rvls.htmlhttp://members.aol.com/johnp71/javastat.htmlhttp://socr.stat.ucla.edu > > good luck, > > bob Hide quoted text  > >  Show quoted text 
Thanks for your reply. As you say, even a large difference of the contribution of a subject to the total between boys and girls does not necessarily imply that one sex actually prefers that subject more. But to move beyond the numbers, we have to, at some point, mke assumptions and draw conclusions from the data. It's never possible to eliminate all lurking variables. All statistical studies can be challenged on some ground. We are trying to do our best within our limited resources. We may be able to increase our sample to include 3 schools from each group, but above that is beyond us. We intend to stick as closely as possible to the data (for example, we will not try to deduce any reasons behind the differences we may see), but it would be rather meaningless if we reported lot of tables without trying to infer some significance. We are hoping that the lurking variables ignored will average out. The consideration of teachers is perhaps not very important, because in Kolkata tutions are increasingly more important than school education.
Is it expected that statistical studies will point out all possible lurking variables ignored?
Also, can you tell me about how to identify different stratas in the population? Thanks.
Molu



