cate;'orios for the latter cases "isere es- 

 tablished when necessary-. 



The v.'ork of all coders was checked by 

 the codinr; supervisor until an acneptable 

 level of coding consistency '..'as achieved 

 both betv.'een and v;ithin coders. There- 

 after a 10 percent check for the purpose 

 of iT.aintaining this consistency level was 

 carried out. 



The punch cards were then prepared 

 and weighted as follows: In Detroit, the 

 interviews completed in sar.plinf; units 

 selected from the white zone were dupli- 

 cated once since these interviewer assifjn- 

 ments had one -half the probability of 

 beinj: included in the sample as did the re- 

 maining sampling units chosen for this 

 sur\'ey in that city. In addition the punch 

 cards for interviev/s com.pleted in assign- 

 ments which had been subsampled were 

 weighted according to the subsanpling 

 rates. No attemprt was made to substitute 

 or v;eight for households desi,:nated for the 

 sample but not interviewed. 



The punch cards then received a thor- 

 ough error and consistency check on the 

 IK-1 Electronic Statistical machine. iJhere 

 necessar>/ the punch cards were corrected 

 by reference to the specific questionnaires 

 corresponding to the cards in question. 



Sampling Krrors 



The sampling error for a particular 

 estimate serves as a guide to the confi- 

 dence with which this estimate can be used. 

 It is a measure of the closeness of the 

 sample estimate to the result v;hich ■'..'ould 



be obtained from a complete census of the 

 population sampled, usin;: the same ques- 

 tionnaire, interviews and interviev;ing 

 procedures. 



Practically all of the estimates 

 developed from the riata collected in this 

 study are simple percentages of the re- 

 spondents having a particular opinion or 

 characteristic. In technical terminology, 

 these percentages are actually combined 

 strata ratio estimates, since the sample 

 design employed extensive geographic 

 stratification and cluster sampling, in 

 which the number of respondents in each 

 cluster was subject to random sampling 

 variation. Thus, sampling errors were 

 computed using the formula for the 

 variance of a ratio estimate. 



This formula contains variance meas- 

 ures for the cluster average of both the 

 numerator and denominator of the computed 

 proportion or percentage estimate, as v:ell 

 as a covariance measure for these two 

 averages. These measures were computed 

 from the average variance between clusters 

 within strata. 



The chances are approximately 2 to 1 

 that the error, due to sampling, in a 

 particular estimate, will not exceed one 

 standard error; the chances are 19 to 1 

 against a deviation as large as two stand- 

 ard errors from the result vjhich v.'ould 

 be obtained with a complete census using 

 the same procedures. 



Estimates of the standard errors for 

 several items included among the guided 

 association questions are shovm in 

 Appendix Table 3. 



hi 



