1 Introduction

Population ageing is occurring throughout the world (United Nations 2017). According to the latest World Population Data, Europe is currently the continent with the largest share (18%) of the population aged 65 or above (World Population Data 2018). Specifically, in 2017, people aged 65+ and 80+ constitute 19.4% and 5.5% of the European Union (EU) population (Eurostat 2018). Similar to many European countries, also Italy has an ageing population. In 2017, older people aged 65+ and 80+ are 22.3% and 6.8% of the Italian population, respectively (our analysis from 2017 Istat data).

Leaving aside the undisputed positive implications of longevity, an aging population presents health systems and society as a whole with new and pressing challenges, linked to the spread of chronic disabilities (Garin et al. 2015). In this context, monitoring older people’s physical and cognitive conditions has become a key issue for policy makers and social researchers. It is therefore not surprising that a number of surveys of older people, especially longitudinal social and health surveys have been conducted in recent decades, both in the U.S. and in Europe (e.g., the Health and Retirement Study in the U. S. and the Survey on Health, Ageing and Retirement in Europe).

Longitudinal surveys are very powerful research resources. However, they pose specific methodological challenges that may hamper the quality of the data collected and undermine the validity of the research findings. In particular, the level of Wave 1 non-response, together with non-participation to the following Waves (i.e. attrition), pose serious challenges to data quality (for an overview, see Lynn 2009); if response is systematically related to outcomes of interest, survey estimates may be biased. This constitutes a serious threat to the generalisability of the research findings.

Fully exploiting the research potential of rich administrative auxiliary data, this paper investigates the individual and household factors associated with Wave 1 response of the longitudinal study Brain Ageing in Abbiategrasso and explores the reasons for refusing to participate in the survey. The paper contributes to the current knowledge in this field in a number of ways. Survey methodology research has disentangled the processes that drive response for surveys of the general population, omitting to address the specificity of the response process in case of different target populations. In our work, we address this specific gap, by focussing on the population of older people. In addition, research in the epidemiology field predominantly focusses on the assessment of the role that respondent characteristics play in influencing response, omitting to consider other factors that may be relevant for survey participation. In our work, we evaluate the extent to which there is a mutual association between two partners’ participation in a survey and systematically investigate gender differences in response pattern.

2 Previous studies on Wave 1 survey participation in longitudinal surveys of older people

To date, there is little research on Wave 1 survey participation in longitudinal surveys of older people, despite the importance of this issue. The lack of knowledge on this topic is documented in many papers; Vind et al. (2009), for example, found that only 43% of all the articles published in high impact journals (and considered in their study) reported Wave 1 response rate, even though this is usually seen as a very basic measure of survey participation. To contextualise our work, we draw on two different, but complementary research fields, i.e. survey methodology and epidemiology.

2.1 Survey methodology research

We are not aware of any studies that performed an in-depth analysis of Wave 1 survey participation in longitudinal surveys of older people. The very few articles in this field provide an overview of Wave 1 response rates for specific surveys, e.g., the Survey of Health, Ageing and Retirement in Europe (SHARE) (de Luca and Peracchi 2005), the English Longitudinal Study of Ageing (ELSA) and the Health and Retirement Study (HRS) (Cheshire et al. 2011). There is also relatively very little research on Wave 1 response in longitudinal household surveys (see, for example, Lynn et al. 2012; Wooden and Watson 2002). The shortage of research in this field is mainly due to the methodological challenges faced when studying Wave 1 non-response, i.e. the lack of auxiliary data on non-respondents (Bethlehem et al. 2011).

To empirically contextualise our study, we consider research on survey participation in cross-sectional household surveys of the general populations, assuming that the Wave 1 response process in longitudinal surveys is similar to the response process in cross-sectional surveys (for an overview of the response process in cross-sectional surveys, see Bethlehem et al. 2011; Groves and Couper 1998, 2012; Stoop et al. 2010). The extensive body of literature in this field has shown that household(er) characteristics are strongly related to response. Specifically, response rates are lowest for both the youngest and the oldest respondents, men, single person households, immigrants and ethnic minority groups, people living in big cities, and respondents at the lowest tail of the income distribution. Survey participation is also positively related to education, occupational social-economic status, being home owner and having small children. Research has also shown that household members’ decisions to participate in surveys are not taken independently; indeed, these decisions are often discussed and negotiated with other members of the household. For example, Sala et al. (2012) documented that household members consult each other and take joint decisions when consenting to data linkage.

2.2 Epidemiology research

The few epidemiological studies that investigated the factors associated with response in surveys of older people performed an assessment of the role played (mainly) by two householder characteristics, i.e. the social and demographic background and physical and mental health. A consistent finding emerging from analysis of these studies is that being/having been married or having/having had a partner is positively associated with response (Adams et al. 1990; Ives et al. 1994; Nummela et al. 2011; Vass et al. 2007; Vind et al. 2009). A number of studies also found that the respondents’ gender and age were unrelated to survey participation (amongst others, Adams et al. 1990; Criqui et al. 1978; Launer et al. 1994; Vass et al. 2007). However, research evidence in this respect is inconsistent; for example, studies by Gao et al. (2015), Gaertner et al. (2016) and Jacomb et al. (2002) found that men and the “youngest” old subjects were more likely to take part in surveys. Similarly, although state of health has been found to be related to response, there is no consensus on the direction of the relationship between these two variables. For example, while Hoeymans et al. (1998) and Knudsen et al. (2010) reported positive associations between health conditions and non-participations in surveys, Van Loon et al. (2003) and Søgaard et al. (2004) found no major differences in self-reported health indicators between respondents and non-respondents.

A minority of studies also explored the role played by socio-economic variables, including education, income and neighbourhood of residence. For example, Vind et al. (2009), Nummela et al. (2011) and Gao et al. (2015) found survey participation to be positively related to the level of income or individual wealth; von Strauss et al. (1998) and Kelfve et al. (2013) have shown that survey response is associated with higher level of education. Although research on Wave 1 survey participation mainly focussed on the study of the associations between specific respondent characteristics and response, some studies also investigated the reasons for refusals. There is some evidence that health (being either too well or too sick) and lack of interest in the study are among the main reasons for non-participation (Adams et al. 1990; Gaertner et al. 2016; Gao et al. 2015; Vass et al. 2007).

In short, our literature review has shown that (1) many household(er) characteristics influence survey participation in cross-sectional household surveys, (2) household members often negotiate survey participation at the household level, (3) marital status is associated with response in surveys of older people, and (4) older people’s health condition and lack of interest in the study are the main reasons for refusals.

3 Research aims

The overall aim of our work is to provide a deeper understanding of the processes underlying Wave 1 survey participation in longitudinal surveys of older people. First, we identify the factors that are associated with response focusing both on individuals’ demographic, health and socio-economic characteristics and the so-called “household contagion effect”, i.e. the extent to which there is a mutual association between two partners’ participation in a survey. Drawing on our literature review, we hypothesise that older people who are married, those with high socio-economic status and older people who live in couples in which the partners take part in the survey are more likely to respond. We do not state any hypothesis concerning the impact of individuals’ health conditions on older people’s survey participation, given the contrasting results in this field.

Second, we explore whether there are any gender differences in older people’s response pattern. Specifically, we state that socio-economic background may play a greater role for older women, who may benefit more than men (e.g., in term of decision making power) from their educational background. In addition, drawing on research on ageing, we know that the loss of a partner may have a worse impact on older men than on older women. In particular, in men the loss of a partner may lead to social exclusion and ill health, including depression and cognitive impairment (Lee et al. 2001; Van Gelder et al. 2006; Van Grootheest et al. 1999). We therefore claim that widows are more likely to participate in surveys than widowers. We do not expect to find any gender differences concerning the impact of older people’s health and the “household contagion effect” on survey response. Third, we also explore the reasons for refusing to respond.

4 Data

We use data from the Brain Ageing in Abbiategrasso study linked to the administrative data from the population registry files (that include also data from non-respondents) provided by the Municipality of Abbiategrasso.

4.1 Brain Ageing in Abbiategrasso study

The Brain Ageing in Abbiategrasso study (also known as InveCe.Ab, ClinicalTrials.gov, NCT01345110) is a biannual cohort study of older people born between 1935 and 1939 (aged 70–75 at the first wave) and living in Abbiategrasso, a town near Milan, in northern Italy. InveCe.Ab is a registry-based population study that aims to assess older people’s physical conditions and identify factors associated with the risk of developing dementia and cognitive impairment. It thus entails the collection of detailed information on respondents’ medical and neuropsychological conditions as well as a wide range of data on their socio-economic conditions, including education, housing, past employment, and social relations. Eligible study participants were identified from local registry office records. Wave 1 was carried out in 2010 and obtained a response rate of 76.3% (the Wave 1 sample members numbered 1773). The Wave 1 data collection was a two-stage process. The first stage included collection of blood samples, administration of a face-to-face questionnaire, and evaluation of walking speed; the second stage involved a medical examination and a neuropsychological assessment of cognitive functioning. Information was also collected on the reasons why eligible individuals refused to participate. The study procedures were in accordance with the principles outlined in the 1964 Declaration of Helsinki and subsequent amendments. The study protocol was submitted to and approved by the Ethics Committee of the University of Pavia. More information on the study can be found in Guaita et al. (2013).

4.2 Population registry files

Italian municipalities maintain population registries that are constantly updated by the Municipality administrative offices. These registries include a wealth of demographic and socio-economic information on Italian citizens, such as year of birth and death, sex, marital status, education, occupation, address, household composition. Unfortunately, they do not collect information on health conditions. We analyse data from the Abbiategrasso administrative offices.

5 Methods of analysis

To perform the analysis of Wave 1 response, we draw on Bethlehem et al. (2011). To identify the variables associated with survey participation and explore any gender differences in Wave 1 response patterns, we ran a set of binomial logistic regressions, reporting the resulting coefficients as odds ratios (ORs). First, we ran a model with main effects and interactions of gender; we then ran the same model (without interaction effects) for men and women separately. To further explore the characteristics of survey participation, we performed bivariate analysis. All analyses were carried out on all potential study participants contacted during the first stage of the Wave 1 data collection process, who comprised 1321 respondents and 323 refusals (129 cases had previously been excluded as non-contactable or ineligible). The statistical analysis was performed using Stata Version 14.0 (STATA Corporation, Texas, USA).

5.1 Variables

The dependent variable in this study was response at the first stage of the Wave 1 data collection process, computed as AAPOR Response Rate 1 (AAPOR 2008). The independent variables were: year of birth, sex, mortality at 24 months (here considered a proxy for general health and calculated considering the occurrence of respondents’ death in the 24 months following the assessment), education, marital status, “household contagion effect”, and neighbourhood of residence - education and neighbourhood of residence are proxies for older people’s socio-economic status. As previously mentioned, all of these variables were derived from population registry files provided by the Municipality of Abbiategrasso and linked to Wave 1 data. Education was measured in years of schooling, while marital status was categorised as married, widowed or single (i.e. divorced and never married). “Household contagion effect” was classified into three categories: both partners took part in Wave 1, only one partner took part in Wave 1, single/no partner eligible for Wave 1. Neighbourhood of residence was coded into four geographical categories that also corresponded to socio-economic differences between the town’s different districts. The town centre is the wealthiest area and the west area is a more rural area; the south-east and north east areas are characterised by a strong presence of white collar and blue collar workers, respectively. It is to be noted that we observed a very low level of item non-response (ranging from 0 to 2.2%) for the variables of interest. The characteristics of the study population are described in Table 1.

Table 1 Characteristics of the InveCe.Ab study population (N = 1644)
Full size table

6 Results

In this section, we discuss the findings of our analysis, focusing on our three research questions.

6.1 What are the variables associated with Wave 1 response?

Table 2 shows the findings from the regression model run on the whole sample. A clear pattern emerges from the analysis of the main effects. Indeed, participation in Wave 1 was found to be associated with most of the variables included in the model. It was positively related to education (OR = 1.16) and also associated with the partner’s decision to take part in the study. In particular, eligible individuals whose partners agreed to participate were nearly four times more likely to take part in the study (OR = 3.90) than those whose partners refused. It emerged that Wave 1 participation was related to marital status and neighbourhood of residence, too. In detail, widowed people were 50% more likely to participate than married people (OR = 1.5), whereas older people living in the West of Abbiategrasso, the most rural part of the town, were about 60% less likely to take part (OR = 0.39). Older people’s health conditions do not seem to affect survey participation. Our hypothesis concerning the associations between older people’s socio-economic status, the household contagion effect and survey response are confirmed. Contrary to our expectations, the hypothesis concerning the relationship between marital status and survey participation is to reject, that is that married people were less likely to participate. Examining two-way interaction effects between our independent variables and gender, we found that the interaction effect between gender and marital status was significant. This is preliminary evidence confirming that older men and women may have different response pattern.

Table 2 Odd ratios (based on logistic regression model) for different characteristics of participants versus non-participants
Full size table

6.2 What were the gender differences in Wave 1 response?

Table 3 shows results from the same regression model (in this case without interactions) run separately for the men and the women. Two clear findings stand out from this analysis; both marital status and neighbourhood of residence showed a variable impact between the sexes.

Table 3 Odd ratios (based on logistic regression models run separately for men and women) for different characteristics of participants versus non-participants
Full size table

Indeed, although there emerged no statistically significant differences when considering women with different marital status (note, nevertheless, that the coefficient for widows was borderline), in the men, on the other hand, marital status was found to be associated with participation in the study; widowers were 44% less likely than married men to take part in the study (OR = 0.56). Furthermore, the effect of neighbourhood of residence disappeared when only the men were considered, whereas it remained significant for the women. In particular, the women from the West were 62% less likely to participate than those living in the centre, which is the wealthiest area of the town (OR = 0.38). Considering neighbourhood of residence as an indicator of social class, we conclude that in the women the decision to take part in the study was partly dependent on their socio-economic background. This analysis also showed some indication of an association between propensity to respond and state of health in older men; in particular, the male respondents who died during the first 24 months of the study were 64% less likely to participate than those who did not (OR = 0.36; p = 0.079). Overall, our hypothesis concerning the role that socio-economic background plays in influencing older men and women’ decision to respond is partially confirmed whereas our expectations regarding the role played by marital status are fully confirmed.

6.3 Why did older people refuse to take part in the survey?

Finally, we investigated the reasons for refusing to participate in the survey. Of the 323 individuals who refused, 63.2% declared that they were not interested in the study, whereas 25.1% and 11.8% did not participate for health-related problems and family commitments, respectively (Table 4). In this older population, we found gender differences in the reasons given for refusal to participate: the men were more likely than the women to refuse out of lack of interest in the study (72.2% vs 58.6%), while the women were more likely than the men to decline because of health-related problems (27.9% vs 19.4%). However, the relationship between the abovementioned variables was significant at the 10% level (Pearson χ2 = 5.7875; p = 0.055; df = 2). The associations between years of education, year of birth, marital status and area of residence and reasons for refusal is not statistically significant (tables not shown).

Table 4 Reasons for refusal by gender (%)
Full size table

7 Conclusions

This study was conducted to explore the Wave 1 response patterns of older men and women participating in the Italian longitudinal study InveCe.Ab and to shed light on the reasons for refusal to participate in the survey. Analysing a set of administrative data, we found that a number of factors are associated with older people’s propensity to participate in surveys, partially confirming our hypothesis. Specifically, our study showed that older people with high socio-economic background and those who lived in couples in which the partners took part conjunctly in the survey were more likely to respond.

However, contrary to our expectation, we found that widowed people were more likely to participate in surveys. In addition, we also documented that older men and women’s pattern of survey participation may differ. Indeed, we found that the impact of older people’s socio-economic background and marital status is different for men and women, because widowers and older women who live in rural areas are less likely to respond. This is a novel result for the epidemiologic research, which so far has underestimated the role that gender plays in survey participation. Taken together, these findings suggest that the responding process has introduced different sources of bias that may have a detrimental impact on the generalisability of the research findings, if not properly taken into account.

Results from our study also indicate that the older people’s response process may have some specificities, i.e. some factors associated with response may play a greater role for older people, compared to the role they play for the general population. Specifically, the household contagion effect seems to play a major role in influencing the partners’ decisions to participate in surveys. Indeed, one of the key findings of this study was that individuals belonging to a couple in which both partners opted to take part in the survey were nearly four times more likely to participate than those in which only one partner agreed. We speculate that this may be due to the fact that older couples have been living together for a long time, sharing similar opinions and adopting analogous behaviours. In addition, as previously discussed, another important feature of older people’s response process is the different role that the socio-economic background and marital status (i.e. widowhood) play for older men and women. Survey designers should consider these specific aspects of older people’s response process, to reduce selectivity in response and limit the number of refusals. For example, interviewers could be trained to tackle the problem of refusals by persuading the most likely member of the couple to participate, and then encourage the other member of the couple to take part, citing their partners’ co-operation; different materials could be prepared to motivate older men and women to respond to the survey (should information about sampling members’ sex be available from the sampling frame).

This study has four major limitations. First, the InveCe.Ab study was carried out in a single town and this inevitably limits the generalisability of the findings. Second, we were not able to explore in depth the influence of health conditions on non-response; this is due to the fact that the administrative data kept by Italian local municipalities do not include this kind of information. Third, selectivity in key substantive survey items, such as respondents’ cognitive abilities, was not assessed, with the result that issues linked to non-response bias in these items remained unexplored. This is because, at the Wave 1 data collection stage, we did not design a shorter version of the questionnaire that could have been used to collect information on individuals who refused to participate. Fourth, the sample size precluded a more detailed analysis of refusals, limiting the statistical and analytical power of the analysis (Chi square test results are dependent on sample size). Further studies are needed to reinforce the findings of this work and shed light on some of the issues that remained unexplored.