3.2 Measures
The outcome of analysis is the participant’s sense of loneliness in ESS Round 6, which was originally measured with the following instrument: ‘How much of the time during the past week you felt lonely?’, and the participant could choose one from the following four options: 1 = ‘None or almost none of the time’, 2 = ‘Some of the time’, 3 = ‘Most of the time’ and 4 = ‘All or almost all of the time’. Following the recommendation by Schneider and Wagemann (
2012: 15) that ‘one should use
fsQCA whenever possible’, the author has calibrated the four values into the following fuzzy membership scores of the set ‘lonely adults’: ‘None or almost none of the time’ = 0.1, ‘Some of the time’ = 0.3, ‘Most of the time’ = 0.7, and ‘All or almost all of the time’ = 0.9.
There are no hard and fast rules for determining the number of causal conditions to be included in a particular set-theoretic analysis, although the general guidance is to keep it ‘at a moderate level’ in order to avoid ‘severe problems of limited diversity’ (Schneider and Wagemann
2012: 276–277). ‘Limited diversity’ and ‘logical remainder’ describe the same problem for QCA, that is, there is insufficient data in a number of rows of the truth table for the analyst to assess the relationship between the configurations and the outcome. Obviously, it is less a problem for a large dataset such as the ESS. In principle, the number of conditions to be included in a large-N study should be determined by striking a balance between the number of possible configurations and their importance in explaining the outcome. On the one hand, the relevance of a condition derived from existing studies is an important criterion for it to be included. In the literature of loneliness, a large number of theoretically important conditions (or risk factors) have been identified, including gender, age, marital status, living arrangement, health conditions and social relations (Anderson
1998; Cacioppo et al.
2010; Prieto-Flores et al.
2011; Victor et al.
2000). On the other hand, as the number of possible configurations (2
k
) increases exponentially with the number of conditions (
k), to include all important conditions will make the analysis overwhelmingly complex, which is why none of the existing large-N QCA studies contains more than five conditions. Therefore, here the number of conditions is limited to five (or 32 possible configurations), making the sample size (2162) more than 67 times of the number of conditions, which is far more than the minimum of four times suggested by Marx and Duşa (
2011).
The rest of this section explains the rationale for selecting and calibrating the conditions in this study. The first is age. Loneliness is widely perceived to be a problem for older people (Arnold-Cathalifaud et al.
2008; Sauer
2006; Tan et al.
2004). Although some researchers have pointed out that loneliness is a serious mental health problem for adolescents as well (Sahin
2012; Storch and Masia-Warner
2004), adolescents were not eligible for participating in the ESS. Note that the question here is whether membership of the set ‘older people’ is a necessary or sufficient condition for feeling lonely, not the statistical association between age and loneliness, as analysed in other studies (Yang and Victor
2011). At what age the word ‘old’ or ‘older’ would be deemed appropriate is highly controversial. Furthermore, as shown in a survey commissioned by the UK’s Department of Work and Pensions (Adams
2013), young people tend to give an earlier age for being old while older people a more advanced one. To be as consistent with public perception as possible, the calibration of age draws on the result from the UK sample of Round 4 of ESS (2008), in which respondents were asked ‘At what age do you think people generally start being described as old?’ The mean is 58.16; therefore, it is sensible to treat those aged below 58 as more out than in the set of ‘old adults.’ In addition, as those aged 80 and above are usually categorized as ‘oldest old’, a score of 0.9 should be appropriate. The calibrated membership scores are presented in Table
1.
Table 1
Calibrations and descriptive statistics
Lonely | All or almost all time | 0.9 | 2.0 |
Most of the time | 0.7 | 5.0 |
Sometimes | 0.3 | 22.8 |
Never or almost never | 0.1 | 70.2 |
Age | ≤29 | 0.1 | 14.2 |
30–39 | 0.2 | 13.9 |
40–49 | 0.3 | 16.3 |
50–57 | 0.4 | 13.6 |
58–64 | 0.6 | 11.9 |
65–70 | 0.7 | 8.5 |
71–79 | 0.8 | 13.0 |
≥80 | 0.9 | 8.6 |
Gender | Female | 1 | 57.6 |
Male | 0 | 42.4 |
Living with spouse/partner | Yes | 1 | 53.6 |
No | 0 | 46.4 |
Self-reported health | Very good | 1 | 27.5 |
Good | 0.8 | 41.0 |
Fair | 0.6 | 22.2 |
Bad | 0.3 | 7.9 |
Very bad | 0.1 | 1.4 |
Frequency of socially meeting with friends, relatives or colleagues | Everyday | 1 | 12.8 |
Several times/week | 0.9 | 29.4 |
Once/week | 0.8 | 22.4 |
Several times/month | 0.6 | 13.5 |
Once/month | 0.4 | 10.1 |
Less than once/month | 0.2 | 9.2 |
Never | 0 | 2.6 |
Besides age, gender and marital status are the other two demographic variables widely included in existing studies on loneliness (de Jong Gierveld
1987; Hawkley and Cacioppo
2010; Victor and Yang
2012). As females are found to be more likely to be lonely than males, ‘gender’ is calibrated into a condition indicating whether a respondent belongs to the crisp set of ‘females’ (1); in the ESS, the only alternative option is ‘male’ (0). Of different types of ‘marital status’, widowhood has been found to be strongly associated with loneliness; however, ‘marital status’ is a multi-value condition, and to study each value as a necessary or sufficient condition would make the analysis over-stretched. In addition, whether an adult lives with their spouse (or partner) logically implies whether the respondent is married or in civil partnership, which is more important than marital status in affecting loneliness. It is therefore sufficient to use ‘living with spouse or partner’ as a much simpler condition. Moreover, in the spirit of QCA, these demographic conditions should not be treated as ‘independent’ or ‘control’ variables, as they are in statistical models; rather, they are analytically on equal footing as other conditions to be specified below.
Physical health is another factor widely identified to be responsible for loneliness. The condition to be included here is ‘self-reported health’, which has five original values: ‘Very good’ (1), ‘Good’ (2), ‘Fair’ (3), ‘Bad’ (4), and ‘Very bad’ (5). As commonly recommended to the application of fsQCA, the threshold or ‘crossover’ value of 0.5 is avoided due to its maximum ambiguity. The interest here is to discover whether a respondent belongs to the set of ‘perceiving oneself as physically healthy’, and it is therefore sensible to treat ‘Fair’ as ‘just in’ with a membership score of 0.6. Accordingly, ‘Good’ has a score of 0.8, and ‘Very good’ has full membership of 1; those in ‘Bad’ health are clearly out of this set (not healthy), therefore are assigned a membership score of 0.3; similarly, those in ‘Very bad’ health have the membership of 0.1.
Finally, a condition that describes the respondent’s social relations must be included. Round 6 of ESS contains three such conditions: ‘How often socially meet with friends, relatives or colleagues’, ‘How many people with whom you can discuss intimate and personal matters’, and ‘Take part in social activities compared to others of same age’. Statistically, they are expected to be strongly correlated and therefore become components of a composite index of sociality. Indeed, the author’s initial plan was to create such a composite index as one condition by summing up the scores of these three variables. It turns out, however, that the level of association among them is not sufficiently strong for such purpose (Cronbach’s alpha = 0.53). That is, these three questions do not seem to be about one latent factor and as a result, each of them must be calibrated individually. However, with four conditions already selected, adding yet three more conditions will make the analysis extremely complicated. In the end, only the first one is included as it has the face validity of measuring the respondent’s level of sociality. The second instrument will be used later as an alternative measure of sociality for testing the robustness of the results.
The frequency of social meetings, originally measured with the values of ‘How often socially meet with friends, relatives or colleagues,’ is taken as the membership of the set of ‘social persons.’ A respondent has full membership (1) if meeting others every day and (0) if never; in between are five fuzzy-set scores: 0.2 for ‘less than once a month’, 0.4 for ‘once a month’, 0.6 for ‘several times a month’, 0.8 for ‘once a week’, and 0.9 for ‘several times a week’.
Sampling weights are not used in QCA as the analysis does not aim to estimate any population parameters with sample statistics. Calibrated memberships of each of the selected conditions with respective percentage are presented in Table
1.