Homework 4 Predictors of Clinically Significant Depression
2025-05-08
Problem Statement
Depression or depressive disorders are classified as a mental disease according to the ICD-11 classification (ICD-11 for Mortality and Morbidity Statistics, 2025). Globally, depression is estimated to have a prevalence of approximately 4%, with women being affected 50% more often than men. Furthermore, it is the fourth most prevalent cause of death among individuals aged 15–29 (World Health Organization, 2023). In the EU, “Spain was the second largest consumer of prescribed anxiolytics […] in 2021” (OECD, 2023, p. 20) with a 6% prevalence of depressive disorders in 2019. A similar gender trend, where females are more affected than men, can also be observed in Spain (OECD, 2023).
From a salutogenic perspective, the question arises: Which social determinants contribute to or impact depression? According to the World Health Organization, social determinants “[…] are the conditions in which people are born, grow, work, live, and age”. These may include working conditions, access to affordable and high-quality healthcare, activities, and age itself (World Health Organization, 2025).
To explore this question, predictor variables (social determinants)
will be identified using data from the European Social Survey, Round 11
(2022). The following hypotheses will be tested for Spain.
1) Depression decreases with age. 2) Frequency of personal internet use
is associated with depression. 3) Volunteering for non-governmental
organisations (NGOs) decreases depression.
Literature
Age as a factor can be examined by analysing depression rates across different age groups. However, this variable may be influenced by several other factors, such as activities, employment, income, or other social determinants. Data on the Spanish population suggests that depression increases with age, showing a left-skewed distribution, with the highest prevalence observed in individuals aged 75–79 (Ministerio de Sanidad, 2021, p. 21).
The relationship between internet usage and depression was studied among older adults in high-income countries (Guo et al., 2025), showing a small reduction in depression (1.4%). Mediating factors such as social interactions, increased physical activity, and access to education helped to mitigate depressive symptoms. Conversely, a cross-sectional study among medical students found that (problematic) internet use led to 27% of students experiencing a depressive episode (Kożybska et al., 2022). These two groups likely differ in their frequency of usage, which will be taken into account in the following analysis.
An umbrella review of Nichol et al. (2024) identified eleven reviews encompassing 41 studies, that examined volunteering as an explanatory variable for depression. 39 of these studies found a positive effect, though moderators such as age and gender influenced the results. Despite the positive effects, volunteering with high empathetic arousals may increase depression. Lorenti et al. (2025) state that volunteering lowers the probability of depression by 5% in the general population and early retired people benefit even more.
Methods
The statistical analysis was performed by using the software R Studio to examine the social determinants of depression in Spain. Depression was treated as the dependent variable, while potential social determinants from the European Survey 11 (ESS 11) (European Social Survey, 2022) were selected as predictor variables.
The construct was operationalised using a short form of the Depression Scale CES, employed in the ESS 11 (2024). The short version (CES-D8) compromises eight items: 1) Felt depressed, how often past week; 2) Felt everything did as effort, how often past week; 3) Sleep was restless, how often past week; 4) Were happy, how often past week; 5) Felt lonely, how often past week; 6) Enjoyed life, how often past week; 7) Felt sad, how often past week; 8) Could not get going, how often past week. All items, except for four and six, are negatively phrased. These two items needed to be reversed in polarity to ensure uniformity in the results. As the scale is ordinal, ranging from “none or almost none of the time” to “all or almost all of the time”, it was converted to numeric values. Similarly, the predictor variables were converted based on their respective scales. To assess the internal consistency of CES-D8, Cronbach’s alpha was calculated. The scale demonstrated multidimensionality and a fair similarity of items (alpha = 0.861).
##
## Cronbach's alpha for the 'dataES[, c("fltdpr", "flteeff", "slprl", "wrhpp", "fltlnl", "enjlf", ' ' "fltsd", "cldgng")]' data-set
##
## Items: 8
## Sample units: 1844
## alpha: 0.487The ESS11 questionnaire was examined for relevant social determinants, which were then confirmed by the literature to formulate hypotheses. To analyse these hypotheses, pairwise associations were applied. For all three hypotheses, Pearson’s correlation coefficient was calculated. Furthermore, a multivariate model incorporating structural variables such as age and gender was developed and supplemented with the pre-analysed variables from the hypotheses.
Results
Sample Description
Gender Distribution
##
## Male Female
## 875 969Age Characteristics
The average age among the Spanish sample is 50 with a minimum age of 16 and a maximum age of 90.
The age distribution of the Spanish population is ought to be seen
below.
Depression Scale
## Item
## 1 Felt depressed, how often past week.
## 2 Felt everything did as effort, how often past week.
## 3 Sleep was restless, how often past week.
## 4 Were happy, how often past week.
## 5 Felt lonely, how often past week.
## 6 Enjoyed life, how often past week.
## 7 Felt sad, how often past week.
## 8 Could not get going, how often past week
## None or almost none of the time Some of the time Most of the time
## 1 59.2 33.1 5.6
## 2 50.4 37.6 8.2
## 3 45.7 34.8 13.3
## 4 31.8 44.7 20.2
## 5 70.5 22.2 4.9
## 6 25.8 43.9 24.0
## 7 44.4 47.7 5.3
## 8 50.7 40.2 6.4
## All or almost all of the time mean count
## 1 2.1 1.5 1844
## 2 3.7 1.7 1842
## 3 6.1 1.8 1843
## 4 3.3 1.9 1843
## 5 2.3 1.4 1843
## 6 6.3 2.1 1842
## 7 2.6 1.7 1844
## 8 2.7 1.6 1843| Item | None or almost none of the time | Some of the time | Most of the time | All or almost all of the time | mean | count |
|---|---|---|---|---|---|---|
| Felt depressed, how often past week. | 59.2 | 33.1 | 5.6 | 2.1 | 1.5 | 1844 |
| Felt everything did as effort, how often past week. | 50.4 | 37.6 | 8.2 | 3.7 | 1.7 | 1842 |
| Sleep was restless, how often past week. | 45.7 | 34.8 | 13.3 | 6.1 | 1.8 | 1843 |
| Were happy, how often past week. | 31.8 | 44.7 | 20.2 | 3.3 | 1.9 | 1843 |
| Felt lonely, how often past week. | 70.5 | 22.2 | 4.9 | 2.3 | 1.4 | 1843 |
| Enjoyed life, how often past week. | 25.8 | 43.9 | 24.0 | 6.3 | 2.1 | 1842 |
| Felt sad, how often past week. | 44.4 | 47.7 | 5.3 | 2.6 | 1.7 | 1844 |
| Could not get going, how often past week | 50.7 | 40.2 | 6.4 | 2.7 | 1.6 | 1843 |
Multivariate Models
The first model incorporated structural variables such as age and gender. The intercept was 1.565 when both variables were included in the depression scale. With each additional year of age, the CES-D8 score increased slightly by 0.001 (p = 0.126), indicating no significant effect. However, gender showed a statistically significant effect, with females scoring 0.178 higher than males (p < 0.001). Adding volunteering and internet use to the model, further decreased the estimate for age (-0.0003). Internet use (est. -0.032; p = 0.004) and volunteering (est. -0.053; p = 0.098) negatively correlated with depression. However, the effect of female gender on depression increased (est. 0.18; p < 0.001).
##
## Call:
## lm(formula = CES_D8 ~ agea + gndr + netusoft_num + volunfp_numeric,
## data = dataES)
##
## Coefficients:
## (Intercept) agea gndrFemale netusoft_num
## 6.259567 -0.002683 1.441439 -0.259860
## volunfp_numeric
## -0.423742##
## Call:
## lm(formula = CES_D8 ~ agea + gndr + netusoft_num + volunfp_numeric,
## data = dataES)
##
## Residuals:
## Min 1Q Median 3Q Max
## -7.2211 -3.1723 -0.8203 2.1620 19.1846
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.259567 0.648067 9.659 < 2e-16 ***
## agea -0.002683 0.006511 -0.412 0.68031
## gndrFemale 1.441439 0.203342 7.089 1.93e-12 ***
## netusoft_num -0.259860 0.090993 -2.856 0.00434 **
## volunfp_numeric -0.423742 0.256224 -1.654 0.09834 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.333 on 1823 degrees of freedom
## (16 observations deleted due to missingness)
## Multiple R-squared: 0.03435, Adjusted R-squared: 0.03223
## F-statistic: 16.21 on 4 and 1823 DF, p-value: 4.692e-13
“To ensure that the sample accurately reflects the larger
population, sampling is applied to minimize bias and increase survey
results reliability and validity of survey results” (survalizer, 2023)
Therefore, the regression model was weighted by using the
analysis function (anweight) to account the effect for Spain.
According to the survey weights’ file of ESS 11, from round nine
onwards, the function is already integrated in the main file. In
comparison to the unweighted model the population baseline is now
higher. Effects of age and internet use were underestimated,i.e. now
have a stronger negative effect on depression. Whereas gender (female)
has remained relatively stable with a slight increase. Volunteering
shows a reduced negative effect after adjustment.
##
## Call:
## lm(formula = CES_D8 ~ agea + gndr + netusoft_num + volunfp_numeric,
## data = dataES, weights = anweight)
##
## Coefficients:
## (Intercept) agea gndrFemale netusoft_num
## 6.59531 -0.00546 1.46096 -0.31612
## volunfp_numeric
## -0.26387##
## Call:
## lm(formula = CES_D8 ~ agea + gndr + netusoft_num + volunfp_numeric,
## data = dataES, weights = anweight)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -11.948 -4.495 -1.198 3.117 29.131
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.595311 0.652271 10.111 < 2e-16 ***
## agea -0.005460 0.006513 -0.838 0.401892
## gndrFemale 1.460962 0.201340 7.256 5.86e-13 ***
## netusoft_num -0.316115 0.092384 -3.422 0.000636 ***
## volunfp_numeric -0.263874 0.253181 -1.042 0.297441
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 6.441 on 1823 degrees of freedom
## (16 observations deleted due to missingness)
## Multiple R-squared: 0.03699, Adjusted R-squared: 0.03488
## F-statistic: 17.51 on 4 and 1823 DF, p-value: 4.161e-14##
## Call:
## lm(formula = CES_D8 ~ agea + gndr + netusoft_num + volunfp_numeric,
## data = dataES, weights = anweight)
##
## Weighted Residuals:
## Min 1Q Median 3Q Max
## -11.948 -4.495 -1.198 3.117 29.131
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.595311 0.652271 10.111 < 2e-16 ***
## agea -0.005460 0.006513 -0.838 0.401892
## gndrFemale 1.460962 0.201340 7.256 5.86e-13 ***
## netusoft_num -0.316115 0.092384 -3.422 0.000636 ***
## volunfp_numeric -0.263874 0.253181 -1.042 0.297441
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 6.441 on 1823 degrees of freedom
## (16 observations deleted due to missingness)
## Multiple R-squared: 0.03699, Adjusted R-squared: 0.03488
## F-statistic: 17.51 on 4 and 1823 DF, p-value: 4.161e-14####Predictors of Clinically Significant Depression The score ranges from 0-24 with a maximum score of 4 per item. Briggs et al. (2018) suggest that a score of 9 and more can be used to identify those with clinically significant symptoms.
##
## 0 1
## 1462 375##
## 0 1
## 0.7958628 0.2041372Females have 1.71 times higher odds of clinical-level depression compared to males, holding other variables constant. Each year of age is associated with a 0.4% increase in odds of depression, but this effect is not statistically significant. Volunteers have 27% lower odds of clinical depression compared to non-volunteers, holding other factors constant. More frequent internet use is weakly associated with lower odds of depression, but this result is not statistically significant.