27.12.2012 Views

The 1995/1996 Household Income, Expenditure - (PDF, 101 mb ...

The 1995/1996 Household Income, Expenditure - (PDF, 101 mb ...

The 1995/1996 Household Income, Expenditure - (PDF, 101 mb ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

1.11<br />

decreases average household sizes. Note also that the unweighted estimates<br />

produce a smaller design effect, meaning that weights raise the sampli¥ error<br />

due to the use of a more complex and less statistically efficient design.<br />

Weighting changes estimate magnitudes in ways small enough to forego<br />

weighting itself, especially given the imprecision of the components which went<br />

into the construction of the weights.<br />

Some of the reluctance to "inflate the sample to national estimates" for<br />

the present study comes from a concern for aggregation bias. For example,<br />

people should not be led to believe that each household's income rises 10%<br />

when GNP rises 10%. Similarly one cannot predict consumption patterns as<br />

national income rises using cross-section Engel curves. Non-linearities enter<br />

into distributions with temporal change. Without complex simulation of the<br />

entire system, microeconomic statistics remain outside the purview of<br />

macroeconomic dynamics.<br />

<strong>The</strong>re is a problem of making comparisons over time. <strong>The</strong> 1990/1991<br />

HIECS sample was only 42.3% rural. CAP MAS sampling staff oversampled in<br />

urban areas in an effort to reduce field costs, and because they believed that<br />

rural areas were "more homogeneous." Oversampling is a fairly common<br />

practice and justified for those reasons, but only if it is followed up by<br />

representative weighting. Unfortunately there is little to be done about this<br />

now because no weights were constructed for the 1990/1991 survey.<br />

Inferences are sensitive to this change in representation. Readers are<br />

strongly cautioned that the effect of different sampling percentages for the<br />

two survey periods becomes more marked the larger the domain of estimation.<br />

One interesting characteristic of the sample selection method for the<br />

<strong>1995</strong>/<strong>1996</strong> HIECS comes from the third stage of sampling, the systematic<br />

selection of 30 households within the sample "chunk." <strong>The</strong>se 30 sample<br />

households are randomly halved so that 15 of the households are enumerated<br />

in the first half of the survey year, then the remaining 15 households are<br />

interviewed in the second half of the year. In other words, all areas are<br />

represented in the first half of the year as well as the second half of the<br />

year. Besides controlling for seasonal effects, such an "interpenetrating"<br />

division permits comparison by half-samples, and any nu<strong>mb</strong>er of sampling and<br />

non-sampling comparisons might be made to detect changes in variability and<br />

bias over the year.<br />

<strong>The</strong> actual target of the enumeration was not the household itself but<br />

the "dwelling unit," the physical space which the household occupied based on<br />

the 1993 listing. <strong>The</strong> enumerator interviewed the present and usual occupants<br />

of the dwelling units, and updated the information about the household head if<br />

the listing was different from the facts on the ground.<br />

LB. Survey Instruments<br />

5 <strong>The</strong> design effect for an estimate is the ratio of the variance of the estimate<br />

given a complex design and the variance of the estimate under a simple<br />

random sample design. In many cases the standard error reported in these<br />

pages assumes a simple random sample, and should be inflated by the square<br />

root of the design effect to obtain the standard error based on the stratified<br />

cluster sample (Le., complex) design.<br />

\\

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!