Mr Heckman confronted the sample-selection problem: in terms of the examples just given, for instance, the effect on the results if the workers who answered the questionnaire differed in important ways from those who did not.
The first step was to figure out the dimensions of theproblem, which required analyzing a sample of Examiner stories and dividing them into four categories: very good or better (20 percent of thesample), good (30 percent), acceptable (20 percent) and poor (30 percent).
Faced with theproblem that DNA gathered at the scene did not match any sample in the national DNA database, they tried what is known as "ancestral DNA profiling".
Before genomics data was offered in the cloud, researchers had to download all the data for a given human sample and discard those portions not applicable to their research problem.