ILO Kish grid

[Sergiy Radyakin]

Dear Statalisters, in the linked ILO document there is a modification of the Kish grid shown in Table 3.1.

It refers to the last digit in the HH questionnaire number as one of the arguments. Digits 1 through 9 are tabulated, but zero is missing.
I don't find any instructions on how the households with last digit 0 (zero) are treated. Have I just missed the instructions and can anyone point me to them, or is it simply "last non-zero digit" instead?

Thank you, Sergiy Radyakin

[Steve Samuels]

Good catch, Sergiy!

The original Kish grid (Kish, 1965, p. 399) had 8 tables; its predecessor, by Deming (1960, p.240) had 12.. Tables are clipped into interviewer folders either in random or (easiest for administration) in fixed order. Linkage of ILO tables to questionnaire digit (skipping the "0" table) was a blunder that should have been caught. However, it's not the worst error.

In the Deming and Kish grids, in a household with K eligible subjects, the subjects are listed from j = 1...K. The grids are constructed so that the probability of selecting the j-th listed subject is very close to 1 in K. So, for example, when there are 2 eligible subjects, the one listed first is selected in half the tables; the one listed second is selected in the others. Selections from list order are effectively at random.

In the ILO tables, youths are to be listed in age order, and the grid is also said to select at random (page 13). It does not. Look, e.g. at the second column, for HH with two youths, the younger is selected in one table, the older in eight. There are severe imbalances and omissions in all the other columns.

My conclusion: the ILO grid was constructed by someone with no understanding of the principles behind Kish's tables. It should never be used.

References:
WE Deming (1960). Sample Design in Business Research, Wiley, NY
L Kish (1965) Survey Sampling, Wiley, NY