Estimates of Sampling Errors for other National Responses 

 or for Sub-S ample s . I'he table gives the sampling error of the 

 national estimate for at least one kind of response to virtually all 

 the questions asked. To estimate approximately what the sampling 

 error for percentages of households giving other responses to the 

 same question would be, consult the tabulation: 



If the response for the And percent for other response to the same 

 sampling error shown is question is: 



this percentage: 



9% or 9$% 



io£ 



or 90% 



20£ 



or 80£ 



30£ 



or 10% 



UO to 60£ 



5% or 10/i 



or 



20)6 



or 



30% 



or 



UO to 60% 



95% 90% 



80* 



error shown 



10% 

 by the 





(Multiply the 



ratio bel 



1.0 l.U 





1.8 





2.1 





2.3 



0.7 1.0 





1.3 





1.5 





1.6 



0.5 0.8 





1.0 





l.i 





1.3 



0.5 0.7 





0.9 





1.0 





1.1 



O.U 0.7 





0.8 





0.9 





1.0 



ow: 



As an example, in the table of sampling errors for the National 

 sample the percent serving canned tuna 1-2 times in past U weeks , (Question 

 0-1) is UU.6, and the standard error of this percent is shown as 1.0. 

 The table containing the results of the household responses to Question 

 1 in Section C, shows the percentage which served tuna U or more times 

 is 2U.5. Assigning UO as the nearest percent given in the table for the 

 percent serving 1-2 times , and 20 as the percent serving U or more times , 

 a ratio of 0.8 is obtained from the above tabulation. This ratio is 

 then applied to the standard error of 1.0 given for 1-2 times , and an 

 estimate of 0.8 percentage points is obtained for U or more times . 



Sampling error for percentage characteristics for subgroups, 

 i.e., percent in a particular region, city- size group, age group, etc., 

 having a particular characteristic, will, of course, be higher than those 

 shown in the table. There is no method by which sampling errors for 

 subgroups can be inferred exactly from the errors for the same character- 

 istics based on the entire sample. However, a rough approximation of 

 the ratio of the two errors can be obtained by dividing the total weighted 

 base by the base used for the particular subgroup, and then taking the 

 square root of the result, ^or exanple, the base for the United States 

 is 2,770 and the base for the Northeast Region is 73U, so that the total 

 base is almost U times the base for this Region. Tailing the square root 

 of U, it TOuld bees tima ted that the sampling error for the Northeast would 

 be about double that given for the sampling error of the national esti- 

 mate of the same characteristic. This ratio would apply to national and 

 regional estimates for the Northeast Region given in Question A-lb and 

 survey results having the same base. 



309 



