In this case, the sample includes too many non-white 

 households. Internal analysis indicates that sampling variability 

 plus somewhat higher return rates for non-white households account 

 for most of this excess, though some difference may be due to the 

 difficulty the interviewer has in classifying Puerto Ricans and 

 Mexicans . 



Refusal rates were somewhat lower for non-whites and this 

 group included fewer chronic away- from- hones, for which the bias is 

 not removed by the special weighting for at/- home frequency. However, 

 at least half the difference seems due to sampling variability. Be- 

 cause of the prevailing segregation pattern, race is a characteristic 

 which has a very high intra-cluster correlation and hence sample esti- 

 mates of racial distributions are subject to more than average error. 



It was decided not to re-weight the data for national 

 estimates to reflect the correct proportions by race, because this 

 re-weighting would have produced only insignificant differences in 

 the results. 



For employment of homemakers, the percentage of respondents 

 (meal-planners) shown by the sample as employed is 29.0. The Bureai 

 of Census figures given in Cur rent Population Reports , Series f-5>0, 

 No. 61, show 29. h percent of housewives (a not exactly comparable group) 

 as employed. 



The income distribution for households is not comparable to 

 any data available from Bureau of Census sources since these are based 

 on family income. The income data are also subject to seme errors of 

 response and non-reporting since in many cases, entries are based on 

 the memory of the housewife rather than on written records. No known 

 data is available by which the accuracy of distributions by religion 

 or by age of the meal-planner can be judged. 



Control Procedures in Processing the Returns 



While field work was still underway, written coding instructions 

 were drawn up for all questions excert the open-end ones. Coding of 

 all but the open-end questions began when the first returns came in 

 from the field. Questionnaires were examined and coded in units of 

 individual clusters. The clusters representing the work of a particular 

 interviewer were grouped together, to permit an early evaluation of the 

 interviewer's ability to carry out the work of the survey. 



313 



