Time trends in the use of field-substitution in the Belgian health interview survey

Study design and use of field-substitution in BHIS

The BHIS is commissioned by both federal and regional authorities. As such, the protocol that governs BHIS stipulates that the data collection must yield a predetermined number of face-to-face interviews in each of the three regions in the country (Flemish Region, Brussels Capital Region and Walloon Region) and, consequently, in an overall fixed net-sample size. BHIS is by design a household survey. Households are invited to participate and when a household participates, at most four household members are selected for participation. To build a probabilistic sample of households, a multistage sampling design is applied, including stratification, clustering, systematic sampling and simple random sampling. Stratification is used to ensure the predetermined number of interviews at regional level [1, 14,15,16].

In both the Flemish and Walloon Regions, provincial stratification is applied, in which the sample size per province is proportional to the population size of the province. The Brussels Capital Region is treated as one province. Within each province, municipalities (Primary Sampling Units, PSUs) are selected through a systematic sampling method with a selection probability proportional to their size. In each selected municipality, systematic sampling (based on statistical sector, age and household size) is used to select households (Secondary Sampling Units, SSUs) using the National Register as the sampling frame. The number of selected households corresponds to fifty individuals who have to be interviewed throughout a 1 year period of data collection (spread over four quarters, so on average 12½ per quarter). Finally, at most four individuals (Tertiary Sampling Units, TSU’s) are invited for participation within each selected household: by definition the household reference person and as appropriate the partner and probabilistic based on a birthday rule two additional household members [1, 14, 17].

Field substitution in the BHIS is applied at the level of the of SSUs: after having hierarchically ordered the households according to statistical sector (smallest territorial unit, comparable with a ward or a district), household size (size 1, 2, 3, and 4+) and age-group (in 7 categories) of the households’ reference person (the person who represents the household in relation with the authorities), per selected household three matched (substitute households) are generated. The so-resulting group of matched quadruples households is called a cluster of households. To account for the uncertainty that the number of interviews cannot be determined with certainty (given possible within household refusal or differences between the administrative and the ‘real-life’ composition of the households selected for participation) for each initial selected cluster a substitute cluster was created that, although composed in exactly the same way, was not matched with the initial cluster for household size and age of the household reference person (1,16,20).

The notion of ‘clusters’ is key for the BHIS substitution procedure. After the clusters have been created, vertical scrambling of clusters and horizontal scrambling of households within the clusters is applied in order to eliminate any kind of systematic approach. At the start of each trimester, all households at the first position of the initial clusters are activated; that is, to all these households an invitation is sent to participate in the BHIS. At the same moment, the interviewers receive the addresses of the households invited for participation, so they can start contacting the households. Each contact attempt has to be registered (day/hour of contact, mode of contact, result of contact) and for each household at least five contact attempts have to be made. In case of frame errors (address does not exist, the residence is clearly abandoned, the household has moved or the address is the address of a prison, a monastery or a psychiatric institution) the household – and all its members - will be labelled as non-eligible and will be substituted. In case the selected households reside at the address and all contact attempts turn out to be fruitless, the household can be labelled as ‘non-contactable’. When a household is contacted, it can refuse to participate (‘refusal’) or it can participate (‘participation’). The latter can only be confirmed once the results of the interview(−s) are uploaded in the central management system. In order to discourage substitution, interviewers are only rewarded for successful participating households. In case a household turns out to be non-contactable or refuses to participate, the central secretariat of BHIS substitutes the household by a household positioned in consecutive order and these households are invited to participate in the BHIS. The procedure to be applied by the interviewers remains unchanged. If necessary, substitution will be applied until the cluster is exhausted (all households of the cluster turn out to be non-contactable or refusing households). Then a substitute cluster will be activated; the household positioned at the first place in the substitute cluster will be invited to participate and will be substituted, if necessary. In the (rare) cases that also the substitute cluster is exhausted, the process ends. The data-collection phase is set at one calendar year, split into four trimesters. Due to field-substitution, data collection is spread throughout the trimesters. If necessary, interviews scheduled for a trimester can be realized in the next trimester(−s). At the end of the year, data collection is interrupted but interviewers have to finalise the interviews in households for which they have appointments. In case the fixed number of interviews in a region is realized before the end of the year, data-collection stops prior to the end of the year [16, 18].

From the first round of BHIS (1997), till the BHIS 2008, data-collection was organized by Sciensano using a interviewers pool specifically created for the survey. From the BHIS 2013 on, data-collection was subcontracted to Statbel, the Belgian Statistical Office and integrated in the data-collection approach used in its other surveys (Labor Force Survey, Statistics on Income and Living Conditions, …).

Data

The number of households invited for participation in the BHIS’s differs according to the survey year, given provincial oversampling (2004, 2008, and 2013), oversampling of the elderly age groups (2004, 2008), oversampling of the German Community (2018) and the protocol-based increase of the base-sample for the Flemish Region (2018). For all households invited for participation in the BHIS’s, info on the survey-year, the region, the age group of the households’ reference person and the household size was derived from the NR used as the sampling frame for BHIS. Initial selected clusters and substitute clusters were identified after vertical scrambling, the position of the households was defined after horizontal scrambling within the (substitute) clusters. The final participation status of the households was derived from the fieldwork management system and linked with NR data. The consistency of the statuses was checked prior to the analysis of the data (e.g. within a cluster a participating household was a household at the first position or could only be preceded by (a) non-participating household).

Analysis

First, the evolution in non-participation rates and field substitution rates will be assessed throughout the survey-years, using an analytic approach inspired by Baldissera et al. [19]: for each of the surveys, initial selected households (INISurvey) lead to participating households (PINISurvey) and non-participating households (NPRINISurvey) (Table 1). Non-participating households are substituted, if needed, several times, by substitute households (SUBSurvey). These substitute households can either participate (PSUBSurvey) or also turn into non-participating households (NPRSUBSurvey).

Table 1 Description of non-response, substitution, participation rates and substitution rates in the Belgian Health Interview Survey

Based on this, the non-participation rates among initial selected households (NPRINSurvey/INISurvey), among substitute households (NPRSUBSurvey/SUBSurvey) and among all households invited for participation (NPRTOTSurvey/TOTSurvey) can be calculated for every survey-year. The substitution rates are expressed as the share of participating substitute households in all participating households (PSUBSurvey/PTOTSurvey), the share of non-participating substitute housholds in all non-participating households (NPRSUBSurvey/NPRTOTSurvey) and the share of all subsitute households invited for participation in all households invited for participation (SUBSurvey/TOTSurvey).

Second, logistic regression is used to assess time trends in substitution, taking into account differences between the successive surveys in terms of age group of the household’s reference person, household size and trimester for which the initial households were invited to participate. Since in BHIS predefined numbers of interviews had to be realized at regional level, the analysis was done stratified by region.

Third, it was examined whether the application of field-substitution changed in terms of the position of the participating household in the clusters (i.e., the probability of finding a participating household in a given survey year at one position in the cluster versus the probability of finding a participating household in another survey year at the same position). These probabilities were calculated from the ESTIMATE statement in the SAS procedure NLMIXED.

留言 (0)

沒有登入
gif