Non-response models for Wave 2
On page 63 of the User Manual, it says that non-response to Wave 2 (for the construction of individual cross sectional weights) is modelled "with covariates taken from responses to UKHLS Wave 2 household grid and household questionnaire".
1. Can you provide any more detail on the specific covariates that were used?
2. Were any individual level covariates used to predict this non-response, given that this was modelled for the construction of individual level weights?
#1 Updated by Dharmi Kapadia about 5 years ago
I also have one further question related to the construction of Wave 2 cross sectional weights. Can you provide more detail on how the longitudinal enumerated individual weight is allocated via the weight share method? Is this based on Lavallee's method (1995)?
#2 Updated by Peter Lynn about 5 years ago
1. These are the variables in the model for households with more than one eligible person (for single-person households, the probability of individual response conditional on household response is de-facto equal to 1.0 at wave 2):
hh2_gor_dv hh2_hhsizep_dv hh2_nkids_dv hh2_npens_dv hh2_nue_dv hh2_nwage_dv hh2_nonepar_dv hh2_tenure_dv hh2_ncars hh2_xfood hh2_xout hh2_xbehind hh2_pcnet hh2_holidays hh2_savings hh2_hsbeds hh2_durab hh2_marstat hh2_employ. If any of these are not self-explanatory, come back to me.
2. Yes, marital status and employment status.
The wave 2 cross-sectional weight was then constructed by multiplying the relevant base weight (e.g. b_psnenus_xw in the case of b_indpxus_xw) by the inverse of the model-predicted propensity of conditional response. The relevant base weight already included some individual-level characteristics in its construction, notably age and sex as recorded at wave 1.
#6 Updated by Dharmi Kapadia almost 5 years ago
Some months ago, I asked about which weight share method was used to construct cross sectional weights at Wave 2. Peter Lynn kindly responded to me, stating it was Lavallee's method.
However, I am having difficulty understanding how Wave 2 cross sectional weights are derived from Wave 2 longitudinal weights. I understand that:
Wave 2 longitudinal weight = Wave 1 cross sectional weight x (1/prob response to Wave 2 conditional on having a non-zero Wave 1 cross sectional weight)
But what is the calculation for the Wave 2 cross-sectional weight. The user guide says that "cross-sectional enumerated individual weights are based on the longitudinal enumerated weights, which are allocated through a weight share method to TSMs and PSMs who entered the sample at Wave 2."
Does this mean that the cross sectional individual weights at Wave 2 are calculated differently for TSMs and PSMs who entered at Wave 2, compared with sample members originating at Wave 1?
Any clarification would be greatly appreciated.
Peter Lynn wrote:
And yes, we used Lavallee's method.
#8 Updated by Peter Lynn almost 5 years ago
The weight-share method treats OSMs, TSMs and PSMs equally. For households containing at least one TSM, the calculation is WC = Sum(WL)/n, where WC is cross-sec wgt, WL is longitudinal wgt, n is number of people (OSMs, TSMs and PSMs) in the household. Note that WL=0 for TSMs.
Thus, for example if the household has one OSM with WL=1.50 and one TSM (with WL=0.00), the weight-share method will give each of those two people WC=0.75.