Response on the network level
Table 2 gives an overview of the willingness of the anchors to participate in the Network Study (see LSNghi055). Of the 1,547 respondents in the main survey, 812 refused (a cooperation of 48%; Schenk et al., 1992 reported for a comparable study 46% cooperation). By using logistic regression, we analyzed whether the anchors in the Network Study form a representative sample of the respondents in the principal sample (the Main Study). The analysis showed that the 812 respondents who refused to take part in the Network Study did not differ from the 735 respondents who - at the time of the face-to-face interview - were willing to cooperate with respect to sex (p= .37), partner status (no partner, living with a partner, partner outside the household) (p= .66), subjective health (p= .70), ADL capacity (p= .74) and network size (p= .11), but differed with respect to age (more younger people than older people were willing to cooperate, p= .004, odds ratio (OR)= .98).
Table 2. Overview of response of the Network Study at the level of the networks
T1
|
T2
|
T3
|
||||||||||||||||
abs. |
% |
abs. |
% |
abs. |
% |
|||||||||||||
| In face-to-face interview asked about network | 4,059 |
|||||||||||||||||
| Not eligible (network size=0) | 15 |
|||||||||||||||||
| Not asked for Network Study (sample) | 2,497 |
|||||||||||||||||
| Eligible for Network Study, assessed at T(x-1) | 1,547 |
100 |
671 |
100 |
594 |
100 |
||||||||||||
| Refusal (*1) | 810 |
52 |
||||||||||||||||
| Refusal before questionnaires were mailed | 64 |
4 |
39 |
6 |
27 |
5 |
||||||||||||
| Anchor deceased | 2 |
0 |
21 |
3 |
21 |
4 |
||||||||||||
| Too little response T(x-1) | - |
- |
15 |
2 |
39 |
7 |
||||||||||||
| Not eligible T(x) (*2) | - |
- |
2 |
0 |
7 |
1 |
||||||||||||
| Questionnaires mailed | 671 |
594 |
500 |
|||||||||||||||
Notes
1 At the time of the face-to-face interview, excluding two respondents who reported after the face-to-face interview that they were willing to cooperate, despite their refusal at the time of the face-to-face interview.
2 A variety of reasons, e.g. all network members have been deceased, severe illness of anchor, networks of two anchors (a couple) overlapped completely and were combined at T3.
Questionnaires were mailed to 671 anchors; two anchors deceased and 64 respondents who were willing to cooperate at the time of the face-to-face interview refused before the questionnaires were mailed. Another nineteen anchors and all of their network members did not return the questionnaires. We thus have at least partial information on 652 networks in the Network Study. A second logistic regression analysis showed that the 3,390 respondents who did not participate in the Network Study did not differ significantly from the 652 respondents who participated with respect to sex (fewer, but not significantly, females in the Network Study, p= .15, OR= .88), subjective health (p= .17, OR= .93), ADL capacity (p= .64, OR= .99), and network size (p= .55, OR= 1.00), but did differ in age (p= .000, OR= .96) and household composition / marital status (p= .000), with about the same proportion living with a partner, (p= .81, OR= .98), more people unmarried and living alone (p= .001, OR= 1.81), about the same proportion divorced and living alone (p= .64, OR= 1.10), about the same proportion widowed and living alone (p= .05, OR= 1.25), and institutionalized (p= .37, OR= .81).
We conclude that our realized sample is biased when we compare it with the Main Study, with the most important deviation for age. However, since the Main Study consists of a stratified sample according to sex and birth year, the realized sample of networks in the Network Study can be viewed as a probability sample.
Response on the level of network members
The selected network members of the remaining 671 respondents (the networks of the two deceased older adults were excluded) and the anchors themselves served as respondents in the Network Study. In the face-to-face interview, the 671 respondents in the Network Study mentioned a total of 4,679 network members who were eligible for the Network Study (the eight network members with the highest frequency of contact). The total number of network members (including anchor) was therefore 5,350 (see LSNghi002). However, when cooperation was requested for the Network Study or in the period between the face-to-face interview and the mailing of the questionnaires, permission to approach the network members was refused by anchor for 1,127 network members (see Table 3).
The decisions by the anchors on who to include in the Network Study were very selective with respect to frequency of contact and type of relationship, but not with respect to sex. By using logistic regression, we analyzed whether the network members (other than the anchors) in the Network Study differ from the network members the anchors mentioned in the face-to-face interview. The analysis (N= 4,651), based on data of the Main Study (LSNa047), shows that about an equal number of males and females were included (p= .03, OR= 1.17), more partners or spouses were included (p= .000, OR= 4.00), more children (p= .000, OR= 2.47), more children-in-law (p= .001, OR= 1.46), about an equal number of brothers of sisters (p= .62, OR= .94), and fewer brothers-in-law and sisters-in-law (p= .000, OR= .51), fewer other relatives (p= .001, OR= .67), fewer friends (p= .000, OR= .67) and fewer neighbors (p= .000, OR= .57); the average frequency of contact (in days per year) between the anchor and the network member was higher for those who were included than for those who were not (p= .01, OR= 1.0008).
For the first wave, 5,306 network members (including the anchors) were eligible. Questionnaires were sent only to network members for whom permission was not refused by anchor at the time of the preparation of the mailing. In addition to the network members for whom permission was refused by anchor at the time of the preparation of the mailing, after the delivery of the questionnaires the anchors refused for a number of network members. The response and non-response for the three waves is shown in Table 3. Note that the figures for the three waves in this table are based on a different number of networks (see Table 2 for an overview of response of the Network Study at the level of the networks).
The response, calculated as proportion of returned questionnaires on the number of questionnaires sent, in the Network Study at T1 was 75% (N= 4,264), and the response was higher for anchors (86%) than for other network members (72%). (We did not correct for the number of network members who were nominated 'double', that is by different respondents and for the number of network members who were respondents of the Main Study, as was done by Schenk et al. (1992). They reported a response rate of 66%.) For T2, the total response was 72% (anchors 78%, network members 71%), and for T3 the total response was 77% (anchors 81%, network members 76%). However, when we calculate the response rates as proportion of returned questionnaires on the number of eligible network members, lower percentages are found (60% for T1, 52% for T2, and 50% for T3). In this calculation is not taken into account that the number of network members differs across networks. We may take this into account by computing firstly the response rate for each network, and averaging secondly these percentages across the networks. The percentages are in that case 61% for T1, 53% for T2, and 51% for T3.
Notes
1 Including four network members who were not among the eight relationships with the highest frequency of contact; they were replacements for other network members.
2 E.g. deceased, ill or left network by movement (e.g. neighbor), retirement of anchor (colleague) or divorce (child in law); assessed at the time of the approach of the network members.
3 Network members other than kin of anchor who did not respond once and network members who did not respond twice were excluded for the next wave.
Response on the level of network relationships
For some analyses, e.g. full network analyses using software like UCINET (Borgatti, Everett & Freeman, 1992), the representativeness of the network data is determined by the number of relationships of which data is available. Table 4 shows the number of relationships available in the Network Study with data (collected from at least one of the two persons within the relationship), and the completeness of data within networks dependent of the criterium applied (see LSNghi055). For example, complete data at T1 are available for 233 networks when we limit our analyses to networks with 80% or more response.
Table 4. Network response (based on the number of relationships)
T1 |
T2 |
T3 |
|
| N networks (*1) | 670 |
592 |
500 |
| M network members | 7.9 |
7.8 |
7.8 |
| Number relationships (*2) | 23,300 |
19,865 |
16,507 |
| Data available (*3) | 14,912 |
10,961 |
8,603 |
| % With data available | 65.9 |
56.7 |
55.4 |
| Number of networks available if we accept | |||
| 70% response | 352 |
239 |
189 |
| 80% response | 296 |
178 |
152 |
| 90% response | 233 |
124 |
94 |
| 100% response | 176 |
81 |
68 |
Notes
1 Non eligible networks excluded
2 These are the eligible relationships of the eight relationships with the highest contact frequency nominated in the Main Study
3 Item non-response may exist
When we compare the sex, age, partner status, subjective health, ADL capacities and network size of respondents of whom the network data are complete with the other respondents with network members, we find in logistic regression analyses for the 70%, 80% and 90% response categories only age to be significant (p< .01, OR about .96) and for the 100% response category age and network size significant (p< .01, both OR about .96). A complete response was more often realized in small networks than in large networks.
Conclusion
Conducting a full Network Study within the framework of a survey is characterized by several obstacles. It is not only a very time-consuming and expensive enterprise, but it is necessary to harm the privacy of anchors and their network members. As a consequence, the response rate is low. However, it is possible to select a relatively large number of networks of which (nearly) complete cross-sectional data are available, and which are representative for the networks of older adults in the Netherlands. For that, one has to select T1 networks with a high but not perfect response. The longitudinal data allows us to study changes within (a large number of) relationships and within (a small number of) networks.
Updated 06-12-2000
