5.3 Same data, different results

For this simple analysis, I use data from Wave 2 of the Panel on Health and Ageing of Singaporean Elderly (PHASE) (see 1.3), conducted in 2011. This is a nationally representative study of older adults aged 60 and above. It is essentially the same dataset used in Lim and Chan (2017) and Chan et al. (2015). Code provided is in R. For the sake of brevity, I leave out observations with any missing values on any of the loneliness items. A key concern of the news article in Figure 5.1 is that even older adults living with family members may be lonely, so we will look at a cross-tabulation of living arrangements with “loneliness”.


Coding scheme 1: Lots of loneliness

I first follow the coding scheme in Lim and Chan (2017) and Chan et al. (2015). I sum the items up (giving me a score that ranges from 0-12), and then dichotomize respondents into people who are “not lonely” (score of 0), and those who are “lonely” (score of 1-12)28. This cut-off point seems to have arisen from a “common-sense” approach rather than any kind of formal testing - that is, group the people who never experience loneliness in one group, and then put the rest who have had some experience of loneliness in another.

# Note: You first need to read in the data
# The data already contains a pre-coded version according to these criteria
lonely1_cat <- phase$w2_loneliness_yesno %>%
  factor(labels=c("Not lonely", "Lonely"))
# Make a table (proportions are weighted to account for survey design)
knitr::kable(
  GDAtools::prop.wtable(livingarr, lonely1_cat, 
                      dir=1, digits=3, w=phase$w2_weights, na=F, mar=F), 
  caption = paste0('Crosstabulation using criteria in Lim and Chan (2017) ',
                   'and Chan et al (2015). Note that these are row percentages.'),
  booktabs = TRUE)
Table 5.1: Crosstabulation using criteria in Lim and Chan (2017) and Chan et al (2015). Note that these are row percentages.
Not lonely Lonely
Living alone 41.022 58.978
Living with spouse only 67.844 32.156
Living with child only 49.910 50.090
Living with spouse and child 71.680 28.320
Living with others only 44.791 55.209


Table 5.1 gives me a similar proportion as suggested in the news article - that is,

“[Associate Professor Chan’s study] found that half of Singaporeans over 60 felt lonely some or most of the time. But those who lived with spouses, or with spouses and children, did not.”


These numbers are indeed worrying. Older adults living alone are understandably lonely, but those who live with their children (but without their spouse) are not that far behind (50.1%!). Even 28% of those who live with their spouse and child feel lonely, like the headline in the news article (Figure 5.1) suggests.


Coding scheme 2: Not that much loneliness

We then arrive at the coding scheme used by Wee et al. (2019) and Ge et al. (2017). Summing the items gives me a score that ranges from 3-9, and I then dichotomize the group into people who are “not lonely” (score of 3-5), and those who are “lonely” (score of 6-9). Note that these cut-points are probably arbitrary - while the researchers cite a paper each to justify their use of the cut-point, the cited papers do not really provide evidence in support of the cut-point. The closest support for the cut-point in the cited papers that I could discern is in Steptoe et al. (2013), which states that they used the top quintile29 to define loneliness. No reason was given as to why the top quintile was chosen. Table 5.2 shows the distribution of “loneliness” according to these criteria.


# Recode and sum the loneliness scores
lonely2 <- phase %>% 
  select(w2_Q10_1_GV1, w2_Q10_2_GV1, w2_Q10_3_GV1) %>%  
  mutate_all(funs(recode(.,`0` = 1, `1` = 1, `2` = 2, `3` = 3, `4` = 3))) %>%
  rowSums()

# Categorize according to cut-off point
lonely2_cat <- if_else(lonely2 < 6, 0, 1) %>% 
  factor(labels=c("Not lonely", "Lonely"))

# Show table
knitr::kable(
  GDAtools::prop.wtable(livingarr, lonely2_cat, 
                      dir=1, digits=3, w=phase$w2_weights, na=F, mar=F), 
  caption = paste0('Crosstabulation using criteria in Wee et al (2019) ',
            'and Ge et al (2017). Note that these are row percentages.'),
  booktabs = TRUE)
Table 5.2: Crosstabulation using criteria in Wee et al (2019) and Ge et al (2017). Note that these are row percentages.
Not lonely Lonely
Living alone 84.419 15.581
Living with spouse only 96.555 3.445
Living with child only 92.184 7.816
Living with spouse and child 97.863 2.137
Living with others only 93.046 6.954


What you will immediately realize is that these numbers are way lower than those when using coding scheme 1 (that is, the coding scheme of Lim and Chan (2017) and Chan et al. (2015)). These numbers are more consistent with the figures shown in Ge et al. (2017)30. Further, the difference in the proportion of those living alone and those living with their child (but without their spouse) is similar in absolute terms, but much larger in relative terms (see Table 5.3). While the proportion of those lonely among those who live alone is 1.2 times that of those living with only their children according to coding scheme 1, this ratio increases to 2 when using coding scheme 2. Based on these results, it seems that the overall loneliness situation is much less dire than before, and those who live with children don’t seem as isolated as those who live alone after all.


Table 5.3: Comparison of absolute and relative differences
Coding scheme 1 Coding scheme 2
(1) Living alone 59.0 15.6
(2) Living with child only 50.1 7.8
Difference [(1) - (2)] 8.9 7.8
Ratio [(1)/(2)] 1.2 2.0


References

Chan, Angelique, Prassanna Raman, Stefan Ma, and Rahul Malhotra. 2015. “Loneliness and All-Cause Mortality in Community-Dwelling Elderly Singaporeans.” Demographic Research S15 (49): 1361–82. https://www.demographic-research.org/special/15/49/.

Ge, Lixia, Chun Wei Yap, Reuben Ong, and Bee Hoon Heng. 2017. “Social Isolation, Loneliness and Their Relationships with Depressive Symptoms: A Population-Based Study.” PLOS ONE 12 (8): e0182145. https://doi.org/10.1371/journal.pone.0182145.

Lim, Ka Keat, and Angelique Chan. 2017. “Association of Loneliness and Healthcare Utilization Among Older Adults in Singapore.” Geriatrics & Gerontology International 17 (11): 1789–98. https://doi.org/10.1111/ggi.12962.

Steptoe, Andrew, Aparna Shankar, Panayotes Demakakos, and Jane Wardle. 2013. “Social Isolation, Loneliness, and All-Cause Mortality in Older Men and Women.” Proceedings of the National Academy of Sciences 110 (15): 5797. https://doi.org/10.1073/pnas.1219686110.

Wee, E. Liang, Y. Tammy Tsang, Huso Yi, A. Sue Toh, L. Geok Lee, Jaime Yee, Shannon Lee, Kellynn Oen, and C. Gerald Koh. 2019. “Loneliness Amongst Low-Socioeconomic Status Elderly Singaporeans and Its Association with Perceptions of the Neighbourhood Environment.” International Journal of Environmental Research and Public Health 16 (6). https://doi.org/10.3390/ijerph16060967.

  1. Note that Chan et al. (2015) further splits the “lonely” group into “sometimes lonely” and “mostly lonely”. Lim and Chan (2017), however, does not make this distinction. I have grouped them together since this is the way that it has usually been represented in public discourse (e.g., the claim that those who are lonely have a higher risk of mortality compared to those not lonely. See, for instance, https://www.straitstimes.com/singapore/those-who-feel-lonely-more-prone)↩︎

  2. In their sample, not the Singapore one.↩︎

  3. Note that the sample in Ge et al. (2017) is of all adults aged 21 and older, not just older adults, so the higher number seen here is expected. Note also that in the paper, the authors show column percentages instead of row percentages. Since we are comparing across living arrangements however, row percentages are more appropriate.↩︎