Why should I not use iweights (in Stata) or WEIGHT BY (in SPSS)?
Computing statistics using iweights (in Stata) or WEIGHT BY (in SPSS) will give the correct point estimate (i.e. the correct value for the mean, proportion or regression coefficient) since the software will use a weighted estimator and thereby remove the known biases that are present in an unweighted statistic. However, the estimator will not correctly compute the variance (and thus the standard error) of the weighted statistic. This is because the software will use the standard SRSWR formula for variance estimation, amending it only to replace n (the number of observations) with Σw (the sum of the weight values for these observations). The true variance is estimated in a much more complex way. A comparison between the use of iweights and the svy procedures in Stata is provided in the sample output under FAQ 5.6. A comparison between the use of WEIGHT BY and use of the Complex Samples module in SPSS provided in the sample output under FAQ 5.7 Another simple way to show that iweights or
