Saturday, December 11, 2010

how to tell if it's a paired.chi-squared, or regression?

Sometimes it is hard to work out whether a question is asking you to do a paired t-test (meaning matched), chi-squared or regression test of a hypothesis.

Here are some identifying tips. Note that I am emphasising how to write the hypotheses. Forcing yourself to write the hypotheses properly is a good way of sorting out what type of test you should do.

matched: look for a column of elements and two columns of data. Each element has a pair of observations, for example ‘before’ and ‘after’. We are trying to work out what difference it makes to the same element. Another example: the one we did in class about Airfares. The element is the destination, and the paired observations are flight to the same place from each of the two airports. Note that there is no information in the elements. It is just a list. Remember how to write the hypotheses?

chi-squared. Now there is some information in the rows of the cross-tabulation. For example, we could have interval-level data here (responses 1 -5 from a survey). Recall that for chi-squared, we are testing for independence between rows and columns. Recall how to write the hypotheses for chi-squared.

regression. The two variables are both quantitative and we are testing for some sort of correlation. If one variable goes up, does the other variable go up (or down)? Recall (very important) the hypotheses for regression. Check the slides for this.

No comments: