There is random error resulting from sampling in an IQ test. For example, some IQ tests have vocabulary sections. How many of the X words the testee knows depends on his/her intelligence but also on which words are chosen.
Because there is random error, if someone took two IQ tests in a short space of time, my default approach would be to average the two scores. If the test dates were taken several years apart, I would give more weight to the more recent score, but I don't know of research saying how much more weight to give it.
IQ is positively correlated with degrees attained, income, and even staying out of jail. For a sample of people who took more than IQ test, it would be interesting to see whether the maximum, the average, or some other function of the test scores best predicted various outcomes.
