Combining Measurements To Yield Generalizable Psychological Functions

Often we wish to measure cognitive performance in a domain in which behavior is strongly and systematically related to some individual difference variable that we are not concerned with. This fact poses two problems from a measurement perspective. First, it adds a source of variability to our sampling distributions. This problem can be annoying and may force us into increasing the sample size of our experiment, but it is hardly fatal. A second, more dangerous, effect is that individual differences may relegate our measurements to a region of parameter space that does not reflect a meaningful or complete range of the behavior in question.

Three general strategies exist to counter the negative effects of individual differences limiting the range of our measurements. First, the researcher can use established theoretical principles in a domain to interpolate or extrapolate to portions of the function that are sparsely occupied by data. Second, the missing data can be inferred statistically by fitting a parsimonious function to the data, such as the lowest order polynomial that accounts for some predetermined proportion of the data. Third, researchers can use a data-collection strategy that ensures sampling across the range of the measurement in question. This can be done by strategically varying the conditions or instructions of an experiment in such a way so as to induce variability along the individual-difference dimension. By doing so, the function relating that dimension to the performance measure can be estimated for each subject. Here I lay out two examples of how this technique is commonly used in memory research. In both of these cases, the solution to the problem of confounding individual differences lies in the elicita-tion of measures across multiple strategically varied conditions.

Was this article helpful?

0 0

Post a comment