Research and analysis

Technical appendix

Published 12 June 2022

Applies to England

1. Technical appendix: Investigating factors associated with loneliness in adults in England

Two datasets were used in this analysis to allow us to address each research question. The Community Life Survey (CLS), which collects data from a random sample of households across England on an annual basis, was used to investigate the relationship between life outcomes and loneliness, and if these relationships have changed over time. Understanding Society, the UK Household Panel study that collects data from the same individuals at multiple time points, was used to investigate the relationships between both protected characteristics and mental health loneliness, and to investigate factors related to the alleviation of loneliness.

All analysis was conducted on data collected from respondents aged 16 and older in England in Understanding Society (USoc) and the Community Life Survey (CLS). CLS analysis was conducted on a sample of 10,243 respondents, and USoc analysis was conducted on a larger sample of 25,494, which lends itself to more complex analytical methods to unpick the predictors of loneliness over time.

The analysis in this report uses survey weights and controls for the complex survey design of both Understanding Society (USoc) and the Community Life Survey (CLS) to ensure that results were representative of adults aged 16 and older living in England. All results presented in this report were formally tested for statistical significance and only presents odds ratios and differences in proportions that were found to be statistically significant at the 95% level. Error bars have been used to illustrate confidence intervals in charts. Results are not statistically significant when error bars overlap.

A range of analytical techniques were used for this report to offer insights into each research question in the most appropriate manner. Details of the analytical methods used in each section of the report are summarised in the table below.

Descriptive statistics

As described in the methods table below, this report employs descriptive statistics to examine the prevalence of loneliness and the relationship between loneliness and a range of factors. This report used proportions – i.e., reporting the percentage of people estimated to be experiencing loneliness in different groups of people – to compare levels of loneliness across key characteristics and experiences.

Repeated cross-sectional proportions have been used to compare these figures when data has been collected from different samples of people at different time points within the CLS.

Logistic regression

The different analytical approaches used in this report were all variations on logistic regression, a technique which allows for the examination of the relationship between a number of explanatory (or predictor) variables and a binary categorical outcome. In this report the outcome was typically a measure of loneliness, and the outcome for each section of the analysis is detailed in the fourth column of the table below. All the logistic regression models discussed below controlled for age, gender, ethnicity, and income level as standard.

Where the table below refers to a ‘logistic regression’ alone, a binary logistic regression with key explanatory and control variables was conducted. However, to better understand the mechanics of loneliness across different groups and time periods, this report also employed three different variations on logistic regression models:

Models estimated for a subsample: Several logistic regression models were conducted on a subsample of respondents to understand the predictors of loneliness for that individual group, e.g. young people, or those already experiencing loneliness.

Models estimated with a lagged predictor: to examine predictors of loneliness over time, several regression models utilised independent variables from a prior wave (e.g. Wave 9 USoc) and an outcome or dependent variable from a later wave (e.g. Wave 10 USoc). Longitudinal weights were employed to ensure representativeness in these instances.

Models estimated with an interaction effect: interactions can be included in logistic regression models to test the joint effect of two variables. To isolate the impact of the joint effect of variables, for all regression models, each model contained no more than one interaction effect, the details of which are specified in the table below.

Report Section	Figure/Table	Survey	Outcome Variable	Analytical Approach
Comparison of loneliness reported in the CLS 2019/20 and Understanding Wave 9	Figure 1, Figure 2	CLS 2019/20USoc Wave 9	Chronic loneliness	Proportions
Chronic loneliness by age and income quintile	Figure 3, Figure 4	USoc Wave 10	Chronic loneliness	Proportions
Day-to-day experiences	N/A	CLS 2019/20	Chronic loneliness	Logistic regression
Factors predicting chronic loneliness over the life course	N/A	USoc Wave 10	Chronic loneliness	Logistic regression
Different risk factors at different life stages: Factors predicting chronic loneliness over the life course	Table 2	USoc Wave 10	Chronic loneliness	Logistic regression: separate model estimated for each age band (16-34, 35-49, 50-64, 65+)
Risk of loneliness for people with protected characteristics	Figure 5	USoc Wave 9	Transient loneliness	Logistic regression: separate models for each interaction between age and gender, age and sex, and sex and gender. One interaction only tested per model.
Wellbeing	Figure 6	CLS 2019/20	Chronic loneliness	Proportions
The impact of past mental distress on loneliness	Figure 7	USoc Wave 9, Usoc Wave 10	Chronic loneliness and mental distress in Wave 10	Logistic regression with lagged independent variables from Wave 9.Conducted on the whole sample, then repeated separately on two sub-samples: 1) those with loneliness at Wave 9, and 2) those without loneliness at Wave 9.
Have the risk factors for loneliness changed over time?	Figure 8	CLS 2016/17 – 2019/20	Chronic loneliness	Repeated cross-sectional proportions, i.e., plotting proportions for the same survey item over successive time periods.
Short term factors that predict the alleviation of loneliness	N/A	USoc Wave 9, USoc Wave 10	Chronic loneliness	Proportions
Short term factors that predict the alleviation of loneliness	Figure 9	USoc Wave 9, USoc Wave10	Chronic loneliness at Wave 10	Logistic regression models conducted for those experiencing chronic loneliness at Wave 9. These models were then estimated for each age group (16-34, 35-49, 50-64, 65+) separately.