This page describes how the following weights were derived for Waves 1-13:
- Cross-sectional weights for each wave
- Longitudinal weights for complete wave sets up to and including each wave from IP2 onwards
- Longitudinal weights for partial wave sets
Common features of all weights
Note that all models used to predict response propensities are fitted using stepwise backward logistic regression with p=0.05. Each set of weights has been scaled by a constant factor to produce a mean of one amongst cases eligible to receive the weight. In consequence, weights which are defined as equal (e.g. d_hhdenip_xw and d_psnenip_xw for the refreshment sample) will not necessarily have the same numeric value, but will retain the same between-person/household relative value.
Wave 1 Cross-Sectional Weights
| a_hhdenip_xw |
| a_psnenip_xw |
| a_indinip_xw |
| a_indpxip_xw |
| a_indscip_xw |
| a_ythscip_xw |
Each IP1 cross-sectional weight consists of a design weight, which is adjusted for non-response and post-stratified to population estimates.
The design weight is the same for a household (a_hhdenip_xd) and for each member of the household (a_psnenip_xd). Design weights are equal for the vast majority of sample members and differ only for cases that involved a) sub-sampling of dwellings because there were more than three dwellings at the address, or b) sub-sampling of households because there were more than three households in the dwelling (see the sample design page).
The first stage of non-response adjustment is at the household level. The adjustment consists of the reciprocal of predicted values from a logistic regression model of household response, where this is defined as completion of at least the household grid (around 59% of households responded in IP1). The covariates in the model were a set of small area indicators and Census 2001 variables, including those used in sample stratification, as well as interviewer observation variables collected during the survey field work. The household weight consists of the design weight with this household-level adjustment. This weight (a_hhdenip_xw) is defined for all households that participated in IP1.
The enumerated person weight (a_psnenip_xw) is defined for all persons in households that participated in IP1. The weight equals the IP1 household weight, post-stratified by age, sex, and grouped Government Office Region. The post-stratification targets are taken from Office for National Statistics 2008 mid-year population statistics. The post-stratification adjustments were calculated for each cell of a 56-cell matrix, as the ratio of population count to weighted sample count. The cells were defined by seven age categories, sex, and four (groups of) regions/countries.
The adult main interview weight (a_indinip_xw) is defined for the 85% of IP1 enumerated persons aged 16 or over who completed the individual interview. It consists of the enumerated person weight, adjusted by the reciprocal of predicted values from a logistic regression model of adult main interview response, conditional on enumeration. The covariates in the model were as described above for household response, plus personal characteristics from the household grid and a limited number of household characteristics from the household grid and household questionnaire. After this non-response adjustment was applied, post-stratification was implemented, based upon the same 56-cell matrix described above for the enumerated person weight.
The adult main or proxy interview weight (a_indpxip_xw) is defined for the 91% of IP1 enumerated persons aged 16 or over for whom either an adult main interview or a proxy interview was completed. It was derived in the same way described above for the adult main weight, the only difference being the definition of response in the logistic regression model.
The adult self-completion weight (a_indscip_xw) is defined for all respondents to the adult main interview who also completed the self-completion questionnaire. (A small number of persons 16 or over completed the self-completion questionnaire but not the adult main interview – these responses are not included in the dataset.) The weight was derived in the same way described above for the adult main weight, but the logistic regression model is based on all enumerated persons aged 10 or over and predicts response to the self-completion (adult or youth) questionnaire.
The youth self-completion weight (a_ythscip_xw) is defined for all persons aged 10 to 15 who completed the youth questionnaire. The weight was derived in an identical way to that described above for the adult self-completion weight.
Wave 2 Longitudinal Weights
| b_psnenip_lw |
| b_indinip_lw |
| b_indpxip_lw |
| b_ythscip_lw |
The IP2 longitudinal enumerated person weight (b_psnenip_lw) is defined for all Original Sample Members (OSMs) enumerated at both IP1 and IP2 – that is, in a responding household at both waves, plus newborns (children of OSM mothers, born between IP1 and IP2). It consists of the IP1 enumerated person weight, adjusted for conditional non-response at IP2. The adjustment consists of the reciprocal of predicted values from a logistic regression model of IP2 enumeration conditional on IP1 enumeration. The covariates were the same as those described above for the IP1 adult main interview weight. Newborns were assigned the same weight as their mother.
The IP2 longitudinal adult main interview weight (b_indinip_lw) is defined for all OSMs who completed the adult main interview at both IP1 and IP2 and for 16 year-old OSMs who completed the interview at IP2 but were too young to be eligible for it at IP1. It consists of the IP1 adult main interview weight (or the IP1 enumerated person weight in the case of IP2 16 year-olds), adjusted for conditional non-response at IP2. The adjustment consists of the reciprocal of predicted values from a logistic regression model of IP2 adult main interview response conditional on IP1 adult main interview response. The covariates were measures from the IP1 adult main interview and the IP1 household grid and household interview, plus interviewer observations.
The IP2 longitudinal adult main or proxy interview weight (b_indpxip_lw) is defined for all OSMs who at both IP1 and IP2 either completed the adult main interview or had a proxy interview carried out on their behalf, and for 16 year-old OSMs who completed either instrument at Wave 2 regardless of their response in IP1 as they were too young to be eligible for the adult main interview at IP1. It should therefore be used in preference to the IP2 longitudinal adult main interview weight for analysis which is restricted to variables that are available from the proxy interview. The weight was derived by adjusting the IP1 adult main or proxy interview weight (or the IP1 enumerated person weight, in the case of IP2 16 year-olds) for non-response at IP2. The adjustment came from a model of IP2 response (main or proxy) conditional on IP1 response (main or proxy). Model covariates were the same as described above for the longitudinal adult main weight, except that variables from the individual interview were restricted to those also included in the proxy interview.
The IP2 longitudinal youth self-completion weight (b_ythscip_lw) is defined for all persons who completed the youth self-completion questionnaire at both IP1 and IP2 and for 10-year-old OSMs who completed it in IP2 but were too young to have been eligible for it at IP1. The weight was derived by adjusting the IP1 youth weight (and the IP1 enumerated person weight, in the case of 10 year-olds at IP2) for non-response to the youth self-completion at IP2. The adjustment came from a model of IP2 youth response conditional on IP1 youth response.
Wave 2 Cross-Sectional Weights
| b_hhdenip_xw |
| b_psnenip_xw |
| b_indinip_xw |
| b_indpxip_xw |
The IP2 cross-sectional enumerated person weight (b_hhdenip_xw ) is defined for all persons (OSMs and Temporary Sample Members (TSMs)) enumerated at IP2. For persons in households where all household members are OSMs, it is equal to the longitudinal enumerated person weight. For persons in households with at least one TSM at Wave 2, it is derived through the weight share method: each person in the household is given a weight of a/b, where a is the sum of the longitudinal enumerated person weights for all OSMs in the household and b is the total number of persons (OSMs and TSMs) in the household.
The IP2 cross-sectional household weight (b_hhdenip_xw) is defined for all households who responded at IP2. It consists simply of the mean[1] of the IP2 cross-sectional enumerated person weights for all persons (OSMs and TSMs) in the household.
The IP2 cross-sectional adult main interview weight (b_indinip_xw) is defined for all persons who completed the IP2 adult main interview (with one exception, noted in the following). In households containing one or more responding TSMs, each respondent in the household is given a weight of a/b, where a is the sum of the longitudinal weights (b_indinip_lw) in the household and b is the total number of respondents in the household who are either an OSM with a non-zero longitudinal weight or a TSM. In all other households (OSM-only households and households in which no TSMs completed the main interview), the cross-sectional weight is equal to the longitudinal weight. Note, that b_indinip_xw will equal zero for all persons in the household if no persons have a non-zero b_indinip_lw. This can happen, for example, if the only person completing the IP2 interview is a TSM or is an OSM who did not complete the IP1 interview.
The IP2 cross-sectional adult main or proxy weight (b_indpxip_xw) is defined for all persons who either completed the adult main interview or had a proxy interview carried out on their behalf. It is derived in a way exactly analogous to that described above for the cross-sectional adult main interview weight, but based upon b_inpxnip_lw instead of b_indinip_lw. Thus, there will be weights of zero for persons in households in which no person has a non-zero value of b_inpxnip_lw.
Wave 3 through 9 Longitudinal Weights for Complete Wave Sets
| n_psnenip_lw |
| n_indinip_lw |
| n_indpxip_lw |
At each wave subsequent to IP2, the basic longitudinal enumerated person weight (n_psnenip_lw) is defined for all persons who were enumerated at all waves up to and including the current one. It consists of the enumerated person longitudinal weight from the previous wave, with an adjustment for nonresponse at the current wave. The adjustment consists of the reciprocal of the predicted probability of enumeration at the current wave conditional on enumeration at all previous waves. The probability is predicted by a model based on OSMs only in which covariates come from the previous wave household grid and household questionnaire. Newborns are assigned the enumerated person longitudinal weight of their biological mother.
At each wave subsequent to IP2, the longitudinal adult main interview weight (n_indinip_lw) is defined for all OSMs who at each wave up to and including the current one completed the interview (or were aged under 16 but were continuously enumerated)[2]. The weight is derived as follows. First, for all OSMs who had a non-zero longitudinal adult main interview weight at the previous wave, a model is fitted to predict completion of the main interview at the current wave. The reciprocal fitted values are multiplied by the longitudinal adult main interview weight from the previous wave. Second, for all OSMs who have reached the age of 16 since the previous wave, reciprocal fitted values from a model of main interview response at current wave conditional on enumeration at current wave are multiplied by the current wave longitudinal enumerated person weight. The model is based on all OSMs aged 16 or over at the current wave, but the predicted values used only for those aged exactly 16. The weights for 16 year-olds are then scaled so that when combined with those for persons aged 17 or over, the weighted proportion of 16-year olds is equal to that obtained by applying the current wave longitudinal enumerated person weight to all enumerated persons aged 16 or over.
Analogously, at each wave subsequent to IP2, the longitudinal adult main or proxy interview weight (e.g. c_indpxip_lw) is defined for all persons for whom either a main or proxy interview was completed at all waves up to and including the current one at which they were eligible (i.e. aged 16 or over). The weight is derived in exactly the same way described above for main interview weights, except that the models predict response to main or proxy interview and that, for persons aged 17 or over, the reciprocal predicted values are multiplied by the previous wave longitudinal adult main or proxy interview weight.
Wave 3 and 4 Longitudinal Weights for Partial Wave Sets
| c_psnenip5_lw |
| d_psnenip13_lw |
| c_indinip5_lw |
| d_indinip13_lw |
| c_indpxip5_lw |
| d_indpxip13_lw c_indscip5_lw d_indscip13_lw |
Longitudinal weights are also provided for the partial wave sets {1, 3}, which is denoted as 5 in the weight variable names, and {1, 3, 4}, which is denoted as 13 in the weight variable names. Note that Wave 2 was carried out using mixed modes and, as a consequence, there are a sizeable number of wave non-respondents at that wave.
These longitudinal weights for partial wave sets were derived using methods directly analogous to those for the longitudinal weights for complete wave sets. For example, the longitudinal enumerated person weight for Waves 1, 3 and 4 (d_psnenip13_lw) consists of the longitudinal enumerated person weight for Waves 1 and 3 (c_psnenip5_lw), with an adjustment for nonresponse at the Wave 4 conditional on response at Waves 1 and 3. For weights that involve separately modelling those who were already eligible at the previous wave in the set and those who have become eligible subsequently (indinip, indpxip, indscip), there will be a larger proportion of sample members defined as newly eligible whenever the previous wave in the set was two or more years previously. For example, creation of c_indinip5_lw (Waves 1 and 3) involved one model for persons already eligible and responding at Wave 1 (to adjust their Wave 1 main interview weight) and another for persons who became eligible for the main interview over the 2-year period between Waves 1 and 3 (to adjust their wave 3 enumerated person weight).
Note that the only longitudinal weights for the adult self-completion questionnaire are for the partial wave sets {1, 3} and {1, 3, 4} as the self-completion questionnaire was not administered at Wave 2.
As with the other longitudinal weights, sample members who had reached age 16 since the previous wave in the set are assigned a longitudinal weight based on their current wave enumerated person weight, adjusted for non-response to the current wave self-completion questionnaire conditional on enumeration.
Wave 3 Cross-Sectional Weights
| c_hhdenip_xw |
| c_psnenip_xw |
| c_indinip_xw |
| c_indpxip_xw |
| c_indscip_xw |
| c_ythscip_xw |
The cross-sectional enumerated person, main interview, and main or proxy interview weights (c_psnenip_xw, c_indinip_xw, c_indpxip_xw) are each based on the respective longitudinal weight c_psnenip5_lw, c_indinip5_lw, c_indpxip5_lw), with application of the weight share method to assign weights to TSMs. In households containing one or more responding (to the relevant instrument) TSMs, each respondent in the household is given a weight of a/b, where a is the sum of the relevant longitudinal weights in the household and b is the total number of respondents in the household who are either an OSM with a non-zero longitudinal weight or a TSM. In other households, the cross-sectional weight equals the respective longitudinal weight. Some respondents will therefore receive a zero cross-sectional weight, namely OSMs with a zero longitudinal weight and TSMs in households containing no OSMs with a non-zero longitudinal weight.
The IP3 household weight (c_hhdenip_xw) equals the minimum of the cross-sectional enumerated person weights (OSMs and TSMs) in the household.
The cross-sectional adult self-completion and youth self-completion weights (c_indscip_xw, c_ythscip_xw) are based on the Wave 3 cross-sectional enumerated person weight, with an adjustment for non-response to the self-completion questionnaire. The adjustment is derived from a single model of response to (either) self-completion questionnaire, based on all enumerated persons aged 10 or over at Wave 3 (who were therefore eligible for either the youth or adult self-completion questionnaire).
Wave 4 Cross-Sectional Weights
| d_hhdenip_xw |
| d_psnenip_xw |
| d_indinip_xw |
| d_indpxip_xw |
| d_indscip_xw |
| d_ythscip_xw |
All IP4 cross-sectional weights combine the continuing sample with the refreshment sample, for whom IP4 was the first wave. Weights were derived in different ways for the two samples.
For the refreshment sample, weights were developed in the same way as for IP1, consisting of a design weight adjusted for non-response. The design weight corrected for multiple dwellings or households at an address. The non-response adjustment is based on covariates from Census 2001 and small area statistics. A separate model was used for each of England, Wales and Scotland as different predictors were available for each country. Adjustments for households in England were based on a model for England only; adjustments in Wales were derived from a model for England and Wales; adjustments in Scotland came from a model for all three countries using only common predictors. (The refreshment sample sizes in Wales and Scotland were too small to support separate models.) These weights serve both as household weights (d_hhdenip_xw) and enumerated person weights (d_psnenip_xw) for the refreshment sample. The enumerated person weight was then post-stratified by gender in London and by gender and five age groups for the rest of Great Britain, based on Office for National Statistics mid-2010 population estimates (the most recent available at the time the weights were derived).
Conditional on enumeration, response to the adult main interview, adult main or proxy interview, adult self-completion and youth self-completion were each modelled using stepwise backward logistic regression. Again, separate models were fitted for England, Scotland and Wales. The covariates used were country-specific and came from neighbourhood statistics, household questionnaire and household grid. The inverse predicted response probability was multiplied by the enumerated person weight to produce the respective weight. Post-stratification was applied for each of the three weights (adult main, adult main or proxy, self-completion – where the latter includes both adults and youth) with the same categories as used for enumerated person post-stratification.
All resulting weights were scaled to a mean of one within refreshment sample.
For the continuing sample, weights were created using the same procedure as for the IP3 cross-sectional weights. The source weights were the respective Wave 1, 3 and 4 longitudinal weights (e.g. d_indinip13_lw in the case of d_indinip_xw). As for IP3, self-completion weights consisted of the cross-sectional enumerated person weight with a non-response adjustment derived from a model of response to the self-completion conditional on enumeration. All resulting weights were scaled to a mean of one within the continuing sample.
Note that no adjustment is made for the fact that the refreshment sample, unlike the continuing sample, in principle includes immigrants to Great Britain since 2008. Under the assumption of ignorable immigration, each of the two weighted samples should represent the 2011 population. Using both together will maximize the statistical precision of cross-sectional population estimates. For this reason each cross-sectional weight as described above is released as a single variable encompassing both samples.
Self Completion Longitudinal Weight Between Waves 5 AND 7
A self-completion weight (e_indscip29_lw) is provided for the longitudinal analysis of Waves 1, 3, 4 and 5. This reflects the omission of self-completion instrument in the second wave of IP. The weight is the product of two parts, the first being self-completion longitudinal weight from Wave 4 (d_indscip13_LW) and the second being an adjustment for nonresponse to the Wave 5 self-completion conditional on self-completion response in all previous waves, but Wave 2. The number 29 reflects the decimal form of the binary representation of the cross-wave response pattern, i.e. 11101 (for Waves 5, 4, 3, 2, and 1 respectively), as described above.
In IP6 and IP7, to reflect that IP2 did not have a self-completion component, a separate weight (f_indscip61_lw, g_indscip125_lw) is created for those panel members who completed the self-completion questionnaire in all eligible waves, i.e. Waves 1, 3, 4, 5, 6 and 1, 3, 4, 5, 6, 7, respectively. These weights were created in the same way as weight e_indscip29_lw, described above, and use this weight as the base.
The self-completion longitudinal weight is discontinued at IP8 to reflect the variety of modes used through the time of IP. The cross-sectional self-completion weight is provided and its calculation is described below.
Longitudinal Weights Incorporating Wave 4 Refreshment Sample
Starting at Wave 5 we provide additional longitudinal weights that include the IP4 refreshment sample. These weights are appropriate for any longitudinal analysis of data collected from Wave 4 (2011) onwards.
The method for adjusting for attrition between Waves 4 and 5 is identical to the one described in the section on longitudinal weights for complete wave sets, with the exception of the base weight. While the weights for complete wave sets do not include refreshment, the weights described here do.
The base weight for enumeration is equal to d_psnenip13_lw, scaled to the mean of 1, for the original part of the sample; and is equal to d_psnenip_xw scaled to the mean of 1 for the 2010 refreshment part of the sample. Conditional on nonzero value for the base weight, the nonresponse between Wave 4 and 5 is modeled. The newborns are assigned the weight of their biological mother. The resulting weight, scaled to the mean of 1, is called e_psneni1_lw.
Similarly, individual response weights e_indpxi1_lw, e_indini1_lw and e_indsci1_lw) are calculated. The base weight is d_indpxip13_lw and d_indinip13_lw for the original sample and d_indpxip_xw and d_indinip_xw for the 2010 refreshment sample (for proxy or main interview and only main interview weights respectively). The attrition correction is identical to the one used for the complete wave set weights, reflecting new base weights. Nonresponse correction for 16 year olds is also identical to the one for complete wave set weights with the exception of the enumeration base weight being e_psneni1_lw.
For IP6, IP7, IP8, and IP9 longitudinal weights are created to enable analysis including the IP4 refreshment sample. The weights are created in an identical way to the longitudinal weights for complete wave sets with the difference in base weight. The base weight is the IP5 longitudinal weight for the combined (original plus IP4 refreshment) sample (e.g. e_psneni1_lw, e_indpxi1_lw, e_indini1_lw or e_indsci1_lw).
Longitudinal Weights from Wave 7 Onwards
We provide a number of different longitudinal weights with a new series of longitudinal weights starting with each refreshment. For longitudinal analysis that starts with IP1 the user should use ‘n_xxxxxip_lw’ weight, for analysis that starts with IP4 (that also includes IP4 refreshment) – ‘n_xxxxxi1_lw’, with IP7 – ‘n_xxxxxi2_lw’, with IP10 – ‘n_xxxxxi3_lw’ and so on.
When a new refreshment sample joins the panel the weight for the refreshment sample is calculated in a similar fashion to the first wave weight. The design weight is set as 1 – reflecting an equal probability sample design. The household nonresponse is then corrected using a number of geographically linked variables to the LSOA codes of the sampled households. The information comes from Census and a number of other government statistics for England and Wales. The nonresponse correction is derived as an inverse of the predicted probabilities from a stepwise logistic regression that predicts enumeration. Because of the small sample size in Scotland we do not run a separate model for Scotland in IP, but give them an average nonresponse correction. Enumerated individuals receive the weight of their household. The household enumeration weight for the refreshment is post-stratified by country, and the person enumeration weight is poststratified by age (5 categories) and sex to match government statistics. This weight is then scaled to the mean of 1 and represents a base weight for the refreshment which is available only for OSM members.
A new refreshment that starts at wave n obtains a longitudinal weight from wave n+1. For this the most inclusive longitudinal weight at wave n that includes all the previous refreshments is joined to the base weight for the refreshment (both are scaled to the mean of 1 beforehand). The person enumeration response at wave n+1 is then modelled conditional on the non-zero value of the joined weight to obtain a new longitudinal person enumeration weight with the subscript ‘im’ (i1 stands for weights that include IP4 refreshment, i2 includes additionally IP7 refreshment, i3 includes IP10 refreshment, and i4 includes IP11 refreshment) . The proxy and individual response weights that include the most recent refreshment are then modelled: the outcome is a proxy (or main) response in both waves n and n+1 (except for the 16-18 year olds who need to respond only in wave n+1) conditional on the longitudinal enumeration weight using predictors from the current wave household grid and household questionnaire. The inverse of the predicted probabilities is then multiplied by the enumeration base weight to create longitudinal weights that include refreshment: n+1_indpxin_lw and n+1_indinin_lw.
Cross-Sectional Weights for Wave 5 Onwards
All IP5 and IP6 cross-sectional weights combine the original (IP1) sample with the IP4 refreshment sample. All IP7, IP8 and IP9 cross-sectional weights combine the original (IP1) sample, IP4 refreshment sample and IP7 refreshment sample. The IP10 cross-sectional weights combine all the previous samples and IP10 refreshment sample. And starting at IP11 cross-sectional weights also incorporate IP11 refreshment in addition to all other samples.
In the wave with a refreshment (e.g. IP7 and IP10) the cross-sectional weight is calculated in two parts, the first part being a ‘usual’ cross-sectional weight (derived from a longitudinal enumeration weight and weight shared to TSMs) for ‘continuing samples’ (all samples excluding refreshment part) and a cross-sectional weight for refreshment part of the sample (which is calculated in the same way as Wave 1 weight, using predictors from linked Census data and other geographical linked data predicting household response, and poststratified by Government Office Region at household level and by gender and 5 age groups at individual level). Both of the resulting weights for ‘continuing’ samples and the refreshment sample are scaled to 1 before being combined into one variable – the cross-sectional enumerated person weight (n_psnenip_xw).
The cross-sectional enumeration person weight (n_psnenip_xw) in waves that do not have a new refreshment subsample (e.g. IP5, IP6, IP8, IP9 and so on) is based on the respective longitudinal weight (n_psneni1_lw between IP5 and IP7; n_psneni2_lw between IP8 and IP10,non-refreshment part of n_psneni3_lw for IP11, and n_psneni4_lw for IP12 onwards), with application of the weight share method to assign weights to TSMs. In households containing one or more enumerated TSMs, each respondent in the household is given a weight of a/b, where a is the sum of the relevant longitudinal weights in the household and b is the total number of respondents in the household who are either an OSM or a TSM. In other households, the cross-sectional weight equals the respective longitudinal weight. Some respondents will therefore receive a zero cross-sectional weight, namely respondents in households containing no OSMs with a non-zero longitudinal weight.
The household weight (n_hhdenip_xw) equals the minimum of the cross-sectional enumerated person weights (OSMs and TSMs) in the household in the waves without refreshment. In the years with refreshment it consists of two parts of which the first one (for continuing subsamples) is equivalent to the household weight in non-refreshment waves, scaled to 1; and the part for refreshment is calculated in the same way as Wave 1 weight, using predictors from linked Census data and other geographical linked data predicting household response, and poststratified by Government Office Region.
The cross-sectional individual response weights for adults (n_indpxip_xw, n_indinip_xw and n_indscip_xw) are calculated conditional on successful enumeration (n_psnenip_xw), and response to the relevant instrument: proxy or main questionnaire, only main questionnaire and self-completion questionnaire.
The cross-sectional weight for youth questionnaire data (n_ythscip_xw) is calculated as the wave n enumerated person weight (n_psnenip_xw) multiplied by an adjustment for nonresponse to the youth questionnaire conditional on enumeration in wave n. Due to the small sample size of youth enumerated sample members, the model uses all respondents age 10 or above and models whether they responded to the self-completion questionnaire excluding predictors that are relevant only for adults (e.g. marriage or employment status). The resulting weight is inferred only to the relevant age group, 10-15 year olds.
Wave 12 Biomarker Weights
For the users who are interested in biomarker data collected in IP12 we provide separate weights. For interviewer or nurse measured estimates of blood pressure please use l_indbpip_xw weight, and those of height and weight please use l_indbmip_xw. If you are interested in combining measured and self-reported blood pressure and/or height and weight measures you could either use a suboptimal weight l_indinip_xw, or create your own weight tailored to your analysis (use l_indinip_xw as your base weight). The weights for measured blood pressure l_indbpip_xw and for height and weight l_indbmip_xw can be found on l_indresp_ip.dta. If you are interested in comparing self-reported measures to those obtained by an interviewer and/or nurse see our online training material Creating tailored weights for UKHLS.
For analyzing full blood samples please use l_indfbip_xw, and for analysing dry blood spots samples please use l_indbsip_xw available on l_indresp_ip.dta. If you are interested in comparing measures obtained from full blood sample to those obtained from dry blood spots sample it is best to restrict your analysis to those who have both measures and use l_indfbip_xw weight. For analysing hair samples please use l_indhsip_xw available on l_hair_ip.dta.
The four weights l_indbpip_xw, l_indbmip_xw, l_indfbip_xw and l_indbsip_xw were calculated using stepwise logistic regression predicting response to blood pressure/ height and weight / full blood samples / dry blood spots samples respectively conditional on positive l_indinip_xw. The predictors used were obtained from IP12 household and personal questionnaires. The reciprocal of the predicted probabilities was multiplied by l_indinip_xw to obtain the final weight. The obtained values were weight shared to all eligible members of a household, and additionally weight shared to those with remaining zero weights. Those who were not eligible for each separate biomeasure were excluded from each model.
Because hair samples were collected from adults and youth the l_indhsip_xw weight is modelled in the same way as above but using l_psnenip_xw as a base weight and predictors from the household questionnaire.
[1] On the main survey, and for subsequent waves of IP, cross-sectional household weights were defined as the minimum of the cross-sectional enumerated person weights, rather than the mean. This was done on the grounds that the probability of a household being enumerated is equal to or greater than the probability of the highest-probability individual in the household being enumerated, so variation between households in the maximum probability may better reflect household probability than variation in the mean probability.
[2] Thus, an OSM who turns 16 and subsequently responds to the adult main interview will have a longitudinal adult main interview weight even though they were not eligible to have a longitudinal adult main interview weight at any previous wave. For example, a sample member who responded to the main interview at IP3 and IP4, but was aged 14 at IP1 and 15 at IP2, will have a non-zero value of d_indinip_lw.
