Scottish Health Survey – telephone survey – August/September 2020: technical report

1.7 Weighting The Data

1.7.1 Introduction

This section presents information on the weighting procedures applied to the survey data. Since 2012, the weighting for SHeS has been undertaken by the Scottish Government rather than the survey contractor (as had previously been the case), but the methodology applied between 2012 and 2019 was largely consistent with that of the 2008 to 2011 sweeps of the survey. The methodology for the 2020 telephone survey was adapted from the 2019 approach to account for the smaller number of completed interviews which took place and to account for some changes in patterns of participants via telephone which are not seen in the face-to-face survey. Full details of the weighting approach is provided in section 1.7.2.

To undertake the calibration weighting the ReGenesees Package for R was used and within this to execute the calibration a raking function was implemented.

1.7.2 Main adult weights

The main adult weight is applicable for all adults interviewed as part of the telephone sample. There are usually six steps to calculating the overall adult weights, but only four actively apply to the telephone weighting. Full details are as follows:

1) Address selection weights (w1)

The address selection weights were calculated to compensate for unequal probabilities of selection of addresses in different survey strata. For the telephone sample there were 32 strata overall (one for each local authority). The address selection weight for each stratum was calculated as:

w1 = Number of PAF addresses in the stratum / Number of addresses selected for the stratum

2) Dwelling unit selection weights (w2)

The Multiple Occupancy Indicator (MOI) for the PAF was used to ensure that if there were multiple dwelling units at a single address point then they would have the same selection probability as individual addresses. However, there are likely to have been some cases where the MOI was incorrect. In face-to-face fieldwork, interviewers record where an MOI is different from PAF when visiting a property. This is not possible via the telephone survey, therefore, the information provided by PAF was assumed to be correct, therefore w2 is effectively 1 for all households.

3) Household selection weights (w3)

Similarly to w2, within a very small number of dwelling units, fieldworkers usually find multiple households, of which only one is selected for participation in the SHeS. Again, due to data collection via the telephone rather than face-to-face, it is not possible to correct for this, therefore w3 was effectively taken as 1 for all households.

4) Calibrated household weights (w4)

The design weight (w1) was applied to the survey data to act as entry weights for the calibration. The execution of the calibration step then modified the entry weights so that the weighted total of all members of responding households matched the population totals for Health Boards, Scotland-level population totals for age/sex breakdown, and the population within each SIMD quintile (20%) areas. The population totals that were used were the NRS mid-2019 estimates for private households. SIMD quintiles were added to the calibration stage rather than the bottom 15% SIMD grouping, which was used in 2019, to try and correct for a change in pattern of participants across deprivation groups which was different to the face-to-face collection.

5) Adult non-response weights (w5)

All adults within selected households were eligible for interview, but within responding households not all individuals completed an interview. The profiles of household members that did not complete the interview were different from those that did. Information on all individuals within responding households was available through information gathered as part of the household interview. This allowed the differential response rates for individuals within households to be modelled using logistic regression to calculate a probability of responding based on their profiles. The logistic regression was only applicable for households containing more than one adult since households consisting of only one adult either responded to the household and individual interviews or did not respond at all. Due to the shorter telephone interview, a lower number of variables was possible to include in this model than in SHeS 2019.

The following variables were considered for inclusion in the model:

• Health Board
• Age/sex
• Number of adults in the household
• Employment status
• Marital status
• Tenure
• Urban/rural classification
• Located within SIMD15 area

Through running backwards and forwards selection procedures for the logistic regression the following variables were included in the final model:

• Health Board
• Age/sex
• Number of adults in the household
• Located within SIMD15 area
• Marital status

The final logistic regression model was then used to calculate the probability of response for all individuals that did respond. The adult non-response weight (w5) was then calculated as the reciprocal of this probability:

w5 = 1 / Probability of individual's response

For households of only one adult the non-response weight was one.

6) Individual calibration and final adult weight (int20wt)

The household (w4) and non-response (w5) were combined (w4*w5) and applied to the survey data prior to the final stage of calibration weighting which matched weighted totals for the survey data to the NRS 2019 mid-year population estimates for Health Boards, age/sex distribution at Scotland level and age/sex distribution for the Glasgow and Greater Clyde Health Board.

Table 2: 2019 mid-year population estimates for private households in Scotland by Health Board

Health Board	Adults	Children	Total
Ayrshire & Arran	303,748	60,961	364,709
Borders	95,603	18,967	114,570
Dumfries & Galloway	123,748	23,175	146,923
Fife	300,086	64,358	364,444
Forth Valley	247,221	52,165	299,386
Grampian	470,618	100,166	570,784
Greater Glasgow & Clyde	954,461	197,711	1,152,172
Highland	262,507	51,621	314,128
Lanarkshire	538,417	118,391	656,808
Lothian	726,147	151,244	877,391
Orkney	18,493	3,536	22,029
Shetland	18,556	4,152	22,708
Tayside	338,949	67,037	405,986
Western Isles	22,115	4,244	26,359
Total	4,420,669	917,728	5,338,397

Total figures might not be exact due to rounding

Table 3: 2019 mid-year population estimates for private households in Scotland by SIMD quintile (20%)

SIMD15	Total population
SIMD Quintile 1 (most deprived)	1,042,416
SIMD Quintile 2	1,042,235
SIMD Quintile 3	1,055,545
SIMD Quintile 4	1,092,833
SIMD Quintile 5 (least deprived)	1,105,368
Total	5,338,397

Total figures might not be exact due to rounding

Table 4: 2019 mid-year population estimates for private households in Scotland by age group

Age group	Male	Female	Total
0-15	470,291	447,437	917,728
16-34	628,520	622,838	1,251,358
35-44	324,887	341,299	666,186
45-54	365,812	392,379	758,191
55-64	355,249	378,471	733,720
65+	459,087	552,127	1,011,214
Total	2,603,846	2,734,551	5,338,397