Return to Project Guidelines
Measurement Error

Measurement error in the Worldwide Governance Indicators is discussed in Party Systems and Country Governance on pages 31-33 and in the associated footnotes.

As described on those pages, the creators of the WGI data estimated that the standard errors ranged from ±0.18 to ±0.23 for five indicators in 2007 and was ±0.26 for the sixth, Political Stability. Typically, standard errors are doubled (i.e., ±0.18 becomes ±0.36) to insure that an observed WGI score is significantly different from the observed score at the standard leve of 95 percent confidence. For this student exercise, we propose the looser criterion of (roughly) one standard deviation, which implies that we are only 68 percent confident that differences between the predicted and observed scores are significant. Relaxing the criterion for this project results in identifying more countries to select for study. We believe that the value of the project comes more from doing the research using countries that interest students.

Apart from measurement error, there is also prediction error, reflected by the standard error of the estimate (SEE), which is the standard deviation of the mean of the squared residuals (deviations around the regression line). If the SEE = 0, the prediction is perfect with no residuals or deviations from predicted values. Thus the SEE incorporates not only measurement error for the variables in the regression equation but also the adequacy of the theory that underlies the regression equation. Not reported in Party Systems and Country Govenance, the standard error of the estimate associated with equation 9.1 is 0.56. Given that the purpose of the research project is to determine underachievers and overachievers based on factors that are not in the equation, using the SEE seems inappropriate.