Closed ld-archer closed 1 year ago
NOTE:: The harmonised ELSA outputs report a binary variable for weekly contact with children, relatives, friends. Shankar et al. have used less than monthly contact for these groups in their social isolation index. We will need therefore to generate these variables from the raw datafiles. Not a problem, but just noting it here in case I find other work that uses the weekly measure and can therefore decide between them. Less than monthly is obviously more extreme, will have to look at the number of people in each category to influence the decision.
Variable has now been included in the model and running without error. Named sociso
for social_isolation (saves some typing).
Lets look at some counts of sociso:
. tab sociso, m
sociso | Freq. Percent Cum.
------------+-----------------------------------
0 | 15,857 15.93 15.93
1 | 17,780 17.86 33.79
2 | 18,375 18.46 52.25
3 | 20,373 20.47 72.72
4 | 18,456 18.54 91.26
5 | 8,696 8.74 100.00
------------+-----------------------------------
Total | 99,537 100.00
It appears that the median value is 3, which is bang in the middle of the index.
. tab sociso wave, m
| wave
sociso | 1 2 3 4 5 6 7 8 9 | Total
-----------+---------------------------------------------------------------------------------------------------+----------
0 | 950 2,762 2,926 2,844 2,109 1,615 1,089 935 627 | 15,857
1 | 1,455 1,571 2,056 1,597 2,036 2,327 2,316 2,064 2,358 | 17,780
2 | 1,851 1,917 2,011 1,494 2,231 2,149 2,264 2,118 2,340 | 18,375
3 | 2,561 2,771 2,604 2,328 2,768 2,103 1,922 1,633 1,683 | 20,373
4 | 3,434 1,933 1,822 2,977 2,067 2,118 1,583 1,315 1,207 | 18,456
5 | 2,097 480 452 1,878 568 1,155 852 693 521 | 8,696
-----------+---------------------------------------------------------------------------------------------------+----------
Total | 12,348 11,434 11,871 13,118 11,779 11,467 10,026 8,758 8,736 | 99,537
. tab sociso educ, m column
+-------------------+
| Key |
|-------------------|
| frequency |
| column percentage |
+-------------------+
| Harmonised Education Level
sociso | 1.Less th 2.Upper S 3.Tertiar . | Total
-----------+--------------------------------------------+----------
0 | 7,595 5,228 1,491 1,543 | 15,857
| 23.50 12.16 9.64 17.65 | 15.93
-----------+--------------------------------------------+----------
1 | 5,360 7,536 3,261 1,623 | 17,780
| 16.58 17.53 21.08 18.57 | 17.86
-----------+--------------------------------------------+----------
2 | 4,568 8,513 3,693 1,601 | 18,375
| 14.13 19.80 23.87 18.32 | 18.46
-----------+--------------------------------------------+----------
3 | 5,802 9,283 3,482 1,806 | 20,373
| 17.95 21.59 22.50 20.66 | 20.47
-----------+--------------------------------------------+----------
4 | 6,059 8,364 2,579 1,454 | 18,456
| 18.74 19.45 16.67 16.64 | 18.54
-----------+--------------------------------------------+----------
5 | 2,942 4,074 967 713 | 8,696
| 9.10 9.47 6.25 8.16 | 8.74
-----------+--------------------------------------------+----------
Total | 32,326 42,998 15,473 8,740 | 99,537
| 100.00 100.00 100.00 100.00 | 100.00
This table shows the number in each category, as well as the column percentages (i.e. proportion of each education group at each level of the index). It shows that there is a complex relationship between education and social isolation in ELSA, with lower education more likely to be associated with a score of 0 on the index, whereas the median values for 2nd and 3rd tier education were 3 & 2 respectively (bit of a twister). Will need to look in the literature to see if this sort of link is what we expect to see.
. tab sociso male, m column
+-------------------+
| Key |
|-------------------|
| frequency |
| column percentage |
+-------------------+
| Male
sociso | 0 1 | Total
-----------+----------------------+----------
1 | 9,330 6,527 | 15,857
| 16.96 14.66 | 15.93
-----------+----------------------+----------
2 | 10,359 7,421 | 17,780
| 18.83 16.67 | 17.86
-----------+----------------------+----------
3 | 11,294 7,081 | 18,375
| 20.53 15.91 | 18.46
-----------+----------------------+----------
4 | 11,308 9,065 | 20,373
| 20.55 20.36 | 20.47
-----------+----------------------+----------
5 | 9,047 9,409 | 18,456
| 16.44 21.14 | 18.54
-----------+----------------------+----------
6 | 3,681 5,015 | 8,696
| 6.69 11.27 | 8.74
-----------+----------------------+----------
Total | 55,019 44,518 | 99,537
| 100.00 100.00 | 100.00
Interesting gender relationships also. Seems women have slightly higher proportions at the lower levels of the index, or women are (on average) less socially isolated than men in ELSA. Would be interesting to see whether there is variation in the elements of the index that explain this. I would expect there to be more women living without a partner due to women outliving their husbands, but there could be some differences in the other elements such as women having more contact with children, or other relatives and friends. This is worth looking into so will set up another issue.
Prediction of the social isolation index looks brilliant from the t-tests. The average of the index as a whole is close to the raw data all the way through, and the prevalences of the individual states are also good (mostly within the 5% tolerance at wave 9, and still very close when not). Overall very happy with this.
Looking again at Shankar et al. (2011), who developed a social isolation index from ELSA to use alongside the revised UCLA loneliness score, they generated an index from the following questions (a positive response to each question adds 1 point to the scale, including each of the subcategories):
The resulting scale ranges from 0-5, with higher scores indicating greater social isolation.
The authors did not consider membership of a gym or sports club, as this is directly related to one of their outcomes (physical activity). Need to think about this as I will probably try to control for this too. Good references in this paper to check out anyway linking the physical activity measures in ELSA with cardiovascular disease mortality (Batty, Shipley, Marmot, & Smith (2003)), future functional impairment (Lang, Guralnik, & Melzer (2007)) and cognitive function (Singh-Manoux, Hillsdon, Brunner, & Marmot (2005)) in older age, all from the Whitehall II prospective cohort study.
ToDo