Response rates

Calculation for response tables

The response rates reported in these tables are designed to measure the effectiveness of fieldwork, and so take as the base those cases that were issued at each wave. This means that cases which were withdrawn before fieldwork started (e.g., if a household contacts us to withdraw from the study in between waves) are not included in the tables. The response rates do not model eligibility for those units with unknown eligibility (e.g., a non-contacted household is counted in the denominator, even though the household may no longer be eligible). The response rates are based on the American Association for Public Opinion Research (AAPOR) standard definitions, RR2 for cross-sectional household response rates and RR1 for the individual response rates. The standard definitions are explained in Section 7 of the 10th edition of the AAPOR  (pages 85-86).

To see how our household and individual response rates were calculated using the interview outcome codes in the data, see below.

Household response

The w_hhsamp file, which includes all issued households, is used to create the household response rates outcome summary variable, using the household outcome variable (w_ivfho).

The household response rate is calculated by using the following formula:

(“Fully responding households” + “Partially responding households” (i.e., at least one adult interview)) DIVIDED BY (All issued households MINUS “Known Ineligible households”)  

The household interview outcome variable (w_ivfho) is used to identify household response statuses as follows:

Codes	Household response status
10 (all eligible hh intv)	code 10 (all eligible hh intv)
11 (interviews + proxies) 12 (interviews + refusal) 14 (hh grid + indiv only (no hh ques)	“Partially responding households” (i.e., at least one adult interview)
13 (hh grid + hh ques only) 39 (lost on laptop)	“Other non-interview”
50 (address not found)	“Untraced mover”
51 (no hh member contact) 52 (unable to locate address) 53 (contact made but not with correct) 54 (address inaccessible) 55 (no phone contact with CATI hhold) 56 (unknown eligibility) 59 (other non-contact)	“Non-contact”
60 (refuse to research centre) 61 (refusal to interviewer) 66 (no web response/web-only hh)	“Refusal”
15 (hh grid + proxies only) 16 (hh grid only) 62 (language problems) 63 (no intv.: age/health) 65 (ill during survey period)	“Other non-interview households”
70-98	Ineligible and not included in the tables

In addition, households with a final interviewer outcome code (w_outcome) of 611 (Office use only: Not issued to interviewer) are excluded from the tables as these represent households that were not worked by the fieldwork agency.

Individual response

The w_indsamp file is used to create the individual outcome summary variable. The individual interview outcome variable (w_ivfio) code is used to identify individuals’ response statuses as follows:

Codes	Household response status
1 (full interview)	“full interview”
2 (proxy interview)	“proxy interview”
9 (lost capi interview) 11 (other non-intvw) 14 (ill/away during survey period) 15 (too infirm/elderly) 16 (language difficulties) 18 (unknown eligibility)	“other non-interview”
10 (refusal)	“refusal”
21 (youth interview) 22 (youth: refusal) 23 (youth: oth non-int) 24 (child under 10) 25 (youth non-interview) 60 (child <16 ref/non-int hh) 61 (chd <16 lang prob/non-int hh) 62 (chd <16 infirm/non-int hh) 63 (chd <16 non-cont/non-int hh) 64 (chd <16 o-o-scope/non-int hh) 67 (chd <16 ill/away – non-int hhold)	All the outcome codes that refer to children are coded as ineligible for the adult response tables.
80 (tsm – no osm/psm) 81 (prev wave adamant refusal) 82 (l-t untrace, w-drawn) 83 (withdrawn before field) 84 (other ineligible) 98 (other retiring (due to health)) 99 (dead)	Ineligible

Other outcome codes that refer to specific outcomes within a non-interviewed household are replaced with their household response variable (see above) and that is used to categorise them as refusals, non-contacts or other non-interview (e.g., the individual outcome is coded as a household non-contact if the household outcome is non-contact).

As above, individuals from households where the final interviewer outcome code is 611 are excluded from the tables.

Effects of mode transition on response rates

In summary, to establish the effects of mode transition during the pandemic, comparisons were made between the 2019 and 2020 samples for the period April to December. This found around three-quarters of those who had completed in CAPI in 2019 took part in 2020 using a different mode. Around one-quarter of those who had not responded in 2019, did respond in 2020. Overall, the response rate for the 2020 sample was just 1.1 percentage points lower than the response in 2019. Response in 2020 was lower among those in the low web propensity sample. Response in 2020 was particularly lower for those in the higher age groups, those who live alone, and those with lower levels of education. An analysis of the unweighted sample composition indicated that there were significant differences in the responding sample in 2020, but that these differences were relatively small, with most under 2 percentage point differences. However, researchers should be aware of the potential for these differences to affect analyses and so use the correct weights or control for factors which may affect response in their models. Further details of these changes are included in the document mentioned below and detailed analyses of mode transition and response behaviour is available in (Alvarez et al.,(2021)).