Suchergebnisse

A study of 14,600 L,'s returned in a survey of buying plans & reading preferences of Consumers Union members finds the incidence of item nonresponse to be widespread. The f of item nonresponse varied with the type of question & with SE characteristics of the R's, such as age, sex, occup, & educ; & was found to increase among questions requiring some thought. or effort on the part of the R. However, the pattern of item nonresponse by R characteristics was very similar for diff types of questions. Somewhat surprisingly, no relationship was found between f of item nonresponse & the position of the question on the 1%. The effect of item nonresponse on parameter estimates is not easily determined. It is clear, however, that item nonresponse can redwe substantially the reliability of estimates. In particular, if allowance is made for possible bias due to this source in estimates of the variances of parameters, the size of the standard errors may be substantially higher than these standard errors in the absence of such an adjustment. An adjustment for such nonresponse would therefore seem highly desirable, not only for estimation purposes but also for planning future operations & allocating resources between measures for reducing sampling errors & means of dealing with nonsampling errors. Methods of adjusting data for item nonresponse are discussed & applied to the data at hand.