Survey Methodology, FY 1995

Data Weighting and Standard Errors of Measurement

Because the survey was conducted on a sample basis for academic institutions, the data were weighted to represent national-level R&D expenditures at institutions of higher education. The survey sample design included four strata of academic institutions. Three strata were certainty strata in which all of the institutions were surveyed. The fourth stratum was a probability stratum in which a sample of institutions that were master's-granting or below was surveyed.

The strata included:

Two of the doctorate-granting institutions (Hahnemann University and Medical College of Pennsylvania) merged during the current survey year.

After survey closeout, sample weighting to universe estimates was performed by increasing the individual questionnaire data values by the inverse of the sampling ratio. Thus, in aggregating the data for institutions from the probability sample for tabulation purposes, each datum value was weighted by the inverse of the sampling ratio.

Estimates derived for institutions in the probability stratum differ to some degree from the numbers that would have been obtained if data were collected for the total population of eligible institutions. The probable magnitude of these differences can be estimated using standard statistical techniques. The relative standard error (coefficient of variation) of an estimate was calculated by dividing the standard error by the estimate. In FY 1995, for total separately budgeted R&D expenditures of $22.1 billion, the standard error of the estimate was $111.1 million at the 95-percent confidence level, with a coefficient of variation of +/- 0.5 percent. Similarly, for the estimate of $13.3 billion in federally financed R&D expenditures, the 95-percent confidence interval was ±$56.7 million, with a coefficient of variation of ±0.4 percent.

[ Survey Population and Sample Design | Data Collection and Processing Activities ]
[ Response Rates and Imputation for Nonresponse | Data Weighting and Standard Errors of Measurement ]