Errata

My article entitled "Replications, Significance Tests, and Confidencein Findings in Survey Research" (POQ 47:261–69) is marredby a serious error which has implications for most of the contentsof the article. Table 1, which is adapted from a table published by Rosenthaland Hall,¹ gives the critical values of required for significance at several levels; itdoes not give the critical values of Trans , or () (). Rather, the latter should be referreddirectly to a table giving the areas under the normal curve,or, in other words, it should be treated as though it were aZ from a single-sample test. This means that the comparisons of significance test resultsfrom pooled data and from series of replications in Tables 2and 3 do not support the conclusion that using a replicate designrather than pooled data lessens the probability of Type II errors.In the case of Table 2, the probability from the pooled sampleis lower than the correct one from the series of replications,and in the case of Table 3, the probabilities are about thesame. The data in Table 4 illustrate a kind of situation in whicha replicate design may be the preferred one for another reason.The replicate design allows an estimate of sampling variabilityfrom the actual variability among a small number of samples,whereas the single-sample and pooled-sample significance testresults may be substantially in error due to the unstable meaningof the "zero" assigned to the reference category for the dummyvariables. However, the estimated probability for the arrayof replications should be .05 rather than the probability reported. Another error with less serious implications for the paper asa whole is that the third "alternative formula" for Trans is incorrect as printed. One of thefirst two formulas, or () (), should be used instead. The major error resulted from a misreading of the paper by Rosenthaland Hall. I apologize to Rosenthal and Hall and to any readersof my paper who have used the table of critical values incorrectlyin their research.