Indice de Precios al Consumidor (IPC) (variaciones porcentuales)
INVERSION EXTRANJERA DIRECTA (Millones de dólares)
4. CARICOM 5 CAN Mercosur
3.5. Análisis A-SEPTE: Metodología y conclusiones
Data weighting proved necessary in the HSE and SHS surveys in order to correct for the fact that, in some cases, the achieved samples did not accurately reflect the population distribution. The weighting strategy in the analyses reported here was informed by the technical reports on the HSE and SHS, and also from detailed correspondence with NatCen undertaken to clarify details of the weighting, and to ensure that the variables used matched our objectives in the analysis of trends across all the survey years.
For the HSE it was felt that the adult sample had provided a good match to the population structure in the majority of survey years. However, a non-response weight was introduced in 2003 and 2004 to adjust for possible biases introduced by participants failing to respond to certain portions of the survey such as the nurse visit where waist hip measurements were taken. However if this weight was included in our analyses, the 2003 and 2004 data would be weighted differently to other years, so it was not used. So, all adult HSE data were unweighted with the exception of 2002, when there was a deliberate over sampling of 16-24 year olds, and 2004 when an adult selection weight was introduced. Where data were not weighted in the original dataset, the generated weight variable takes the value 1. In keeping with this, to preserve the original sample size, all weight variables taken from the original data were scaled to have mean 1 (by dividing the weight variable by its mean). It was unfortunately not possible to combine the extra care home sample in 2000 or ethnic boost samples in 1999 and 2004 into the main survey
sample, as despite considerable correspondence on the topic with NatCen, a suitable weight did not exist to correct the distributions.
Child data required weighting in all survey years, as the number of children selected from a household was limited to a maximum (the sampled fraction of children therefore being smaller in households with more than this number of children). An inverse probability (of selection) child weight was used, which may be thought of as the total number of children in a household divided by the number of selected children from that household. In some years further small changes were made to this weight to further correct age and sex distributions.
We now consider how the weights were built up by year for the HSE:
• 1991 to 1994: children were not present in the sample and all data remained unweighted.
• 1995, 1996, 1998 & 2001: the child weight was created from a single child selection weight variable.
• 1997: a number of possible child weights were created by NatCen for different analysis purposes and the selected weight was the weight that maintained the effective sample size.
• 1999: data were divided into a general population sample, which merely had a child weight, and a separate ethnic minority boosted sample which collected data from the various ethnic minorities within the UK. All the data from this second sample was weighted (including children) in order to correct for non-equal chances of selection in the ethnic sub-population. However NatCen advised that there was no weight available which would allow an analysis combining these two sub-samples so here only the general population sample was retained.
• 2000: only the general population sample has been retained as there was no weight available to combine the care home sample data with the general population; again this general
population was weighted only for children.
• 2002: a selection weight was applied to both adults and children, correcting for the over sampling of adults under 25 and mothers with a child of age less than 1 year, and the usual child selection weight.
• 2003: There were selection weights available for children but not adults. Non-response weights were introduced for adults and children for the interview and nurse visit stages, but
analysis. The analysis in this report therefore used unweighted adult data and only the child selection weights
• 2004: HSE had derived weights for both adults and children that combined selection weights and non-response weights for nurse visits. It was desirable to only utilise the component which corresponded to selection weights, for comparability with previous surveys. NatCen were able to provide us with a selection weight for adults, but although requested, did not do so for children. The child data were therefore analysed using the combined weights. The SHS followed a slightly different weighting approach, in that, in both 1995 and 1998 all available data (no children were sampled in 1995) was been given a selection weight in order to correct for non-equal probabilities of inclusion in the survey depending on factors such as postcode. In 2003 the approach largely mirrors that of the HSE with the use of a combined weight which encompassed selection and non-response weights in both adults and children. It was desirable to only utilise the component which corresponded to selection weights, for comparability with previous surveys. However although the calculation of selection weights only was requested from NatCen, they were not provided and the analysis used the combined weights.