FINAL PAPER- revised according to Likert, weights and readability
Determinants of depression in Italy: individual, interpersonal and social factors.
Battisti Sofia
2025-05-15
1. INTRODUCTION
In recent decades, mental health has emerged as a
critical public health concern globally.
According to Wu et
al. (2023), the prevalence of mental health disorders
has increased substantially between 1990 and
2019, and projections indicate a continued upward trend in the
coming years.
This alarming trajectory underscores the need to
deepen our understanding of the social, psychological, and environmental
factors that contribute to mental health outcomes across different
populations.
This paper aims to contribute to the existing body of mental health
research by exploring a set of psychosocial and contextual
determinants that might influence depressive symptoms among the
Italian population. Specifically, the study investigates whether
individuals’ perceived control over their lives, overall satisfaction
with life, interpersonal trust, confidence in the health system, and
satisfaction with governmental efforts to address climate change are
associated with variations in depressive symptomatology.
The analysis is based on data from the European Social Survey
(ESS11, 2023), a cross-national survey that provides
comparative data on attitudes, beliefs, and behavioral patterns across
European countries.
By focusing on the Italian
subsample, this research seeks to identify key
predictors of mental health and offer insights that drive
health interventions and public policy strategies aimed at mitigating
the burden of depression in Italy.
2. LITERATURE REVIEW
Depression is a multifactorial condition influenced by individual,
interpersonal, and societal variables. This study hypothesizes that
greater perceived life control and satisfaction, along with higher trust
in others and societal institutions, are associated with reduced
depressive symptoms.
Nguyen et al. (2020) found that perceived life control
mitigates the effects of external stressors more effectively
than trust in institutions or religious reliance. Similarly,
life satisfaction plays a preventative role against
depression, with a moderately bidirectional relationship to
perceived control (Zalewska et al., 2021).
Interpersonal trust also emerges as a key protective factor. Martinez
et al. (2019) and Zhang (2024) show that strong social
ties—especially with family and neighbors—enhance
emotional support and resilience.
Trust in healthcare systems further
influences mental health outcomes; individuals who
perceive healthcare as reliable are more likely to seek support,
reducing the risk of worsening symptoms (Ahnquist et al., 2010;
Rasanathan, 2024).
Environmental concerns also intersect with mental health. Shen et
al. (2024) demonstrate that effective climate policies,
such as carbon trading, can positively impact psychological
well-being, especially in vulnerable rural populations.
Collectively, these findings underline the need for a holistic and multisectoral approach to understanding depression, emphasizing the interplay of personal agency, social trust, and institutional confidence.
3. METHODOLOGY
1 After extracting data related to the Italian
country, I created the CES_D8 Depression Scale.
This scale is based on d20-d27 variables from the ESS11 survey.
Happiness, sadness, loneliness, joy and depressive feelings, as well as
sleeping habits are combined to create the dependent variable CES_D8
that evaluates personal wellbeing from different points of view.
First of all, creating the CES_D8 Scale requires to check the polarity of the chosen variables.
DataIT$wrhpp <- factor(DataIT$wrhpp, levels = rev(levels(DataIT$wrhpp)))
DataIT$enjlf <- factor(DataIT$enjlf, levels = rev(levels(DataIT$enjlf)))After having changed the polarity of “happiness” and “joy”, I need to convert the other variables into numeric ones, so that the final scale can be calculated.
DataIT$fltdpr_num <- as.numeric(DataIT$fltdpr)
DataIT$flteeff_num <- as.numeric(DataIT$flteeff)
DataIT$slprl_num <- as.numeric(DataIT$slprl)
DataIT$wrhpp_num <- as.numeric(DataIT$wrhpp)
DataIT$fltlnl_num <- as.numeric(DataIT$fltlnl)
DataIT$enjlf_num <- as.numeric(DataIT$enjlf)
DataIT$fltsd_num <- as.numeric(DataIT$fltsd)
DataIT$cldgng_num <- as.numeric(DataIT$cldgng)Now that all variables are numeric, the CES_D8 Scale can be computed by doing the rows’ sum.
DataIT$CES_D8 <- rowSums(DataIT[, c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num", "fltlnl_num", "enjlf_num", "fltsd_num", "cldgng_num")])-8
library(knitr)
kable(t(summary(DataIT$CES_D8)),
caption = "Summary Statistics for CES-D8 Score",
digits = 2)| Min. | 1st Qu. | Median | Mean | 3rd Qu. | Max. | NA’s |
|---|---|---|---|---|---|---|
| 0 | 3 | 5 | 6.09 | 8 | 24 | 100 |
At this point, before checking the Scale’s reliability, it would be
interesting to understand how people answered to the single items that
compose our CES_D8 Scale.
This, is going to make clear how many
people have, distinctively, feel lonely, sad, depressed and so on.
library(likert)## Loading required package: ggplot2## Loading required package: xtablelibrary(knitr)
library(kableExtra)vnames <- c("fltdpr", "flteeff", "slprl", "wrhpp", "fltlnl", "enjlf", "fltsd", "cldgng")
likert_df <- DataIT[, vnames]
likert_table <- likert(likert_df)
plot(likert_table, main = "Likert Plot for CES-D8 Scale", xlab = "Percentage")
This chart correctly visualizes respondents’ answers. However, it might be interesting to quantify such answers but now the mean and counts of the survey’s responses allow us to create a numeric, quantifiable overview of such responses.
likert_numeric_df <- as.data.frame(lapply(DataIT[, vnames], as.numeric))
likert_means <- sapply(likert_numeric_df, mean, na.rm = TRUE)
likert_counts <- sapply(likert_numeric_df, function(x) sum(!is.na(x)))
likert_table$results$Mean <- unlist(likert_means)
likert_table$results$Count <- unlist(likert_counts)
likert_table$results$Item <- c(fltdpr = "Felt depressed", flteeff = "Felt everything was an effort", slprl = "Sleep was restless", wrhpp = "Felt happy", fltlnl = "Felt lonely", enjlf = "Enjoyed life", fltsd = "Felt sad", cldgng = "Could not get going")
likert_table$results[, 2:6] <- round(likert_table$results[, 2:6], 1)
likert_table$results$Mean <- round(likert_table$results$Mean, 3)
kable(likert_table$results, caption = "Distribution of CES-D8 Scale Responses (Italy)", digits = 2) %>%
kable_styling(full_width = F)| Item | None or almost none of the time | Some of the time | Most of the time | All or almost all of the time | Mean | Count |
|---|---|---|---|---|---|---|
| Felt depressed | 71.1 | 24.0 | 4.1 | 0.8 | 1.3 | 2840 |
| Felt everything was an effort | 52.6 | 37.4 | 7.3 | 2.7 | 1.6 | 2838 |
| Sleep was restless | 53.7 | 38.1 | 6.0 | 2.2 | 1.6 | 2850 |
| Felt happy | 14.8 | 43.3 | 34.2 | 7.7 | 2.3 | 2817 |
| Felt lonely | 61.8 | 29.6 | 5.9 | 2.7 | 1.5 | 2837 |
| Enjoyed life | 9.8 | 29.7 | 45.7 | 14.7 | 2.7 | 2803 |
| Felt sad | 48.3 | 45.3 | 4.5 | 1.9 | 1.6 | 2832 |
| Could not get going | 57.2 | 35.4 | 5.7 | 1.7 | 1.5 | 2825 |
2 In this second step, the Cronbach’s
Alpha will be employed. The goal is to check the Scale’s
reliability, as this is required step when dealing with composite
scores.
1. To correctly measure the Cronbach’s Alpha, the variables
values will be checked, as the presence of NA’s could then affect the
computation’s reliability.
library(knitr)
vars <- c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num", "fltlnl_num", "enjlf_num", "fltsd_num", "cldgng_num")
summary_output <- summary(DataIT[, vars])
summary_df <- as.data.frame(matrix(summary_output, ncol = length(vars)))
colnames(summary_df) <- vars
summary_df$Statistic <- c("Min", "1st Qu.", "Median", "Mean", "3rd Qu.", "Max", "NA's")
summary_df <- summary_df[, c("Statistic", vars)]
kable(summary_df, caption = "Summary Statistics of Selected Variables", digits = 3)| Statistic | fltdpr_num | flteeff_num | slprl_num | wrhpp_num | fltlnl_num | enjlf_num | fltsd_num | cldgng_num |
|---|---|---|---|---|---|---|---|---|
| Min | Min. :1.000 | Min. :1.000 | Min. :1.000 | Min. :1.000 | Min. :1.000 | Min. :1.000 | Min. :1.0 | Min. :1.000 |
| 1st Qu. | 1st Qu.:1.000 | 1st Qu.:1.000 | 1st Qu.:1.000 | 1st Qu.:2.000 | 1st Qu.:1.000 | 1st Qu.:2.000 | 1st Qu.:1.0 | 1st Qu.:1.000 |
| Median | Median :1.000 | Median :1.000 | Median :1.000 | Median :2.000 | Median :1.000 | Median :3.000 | Median :2.0 | Median :1.000 |
| Mean | Mean :1.347 | Mean :1.601 | Mean :1.567 | Mean :2.348 | Mean :1.497 | Mean :2.654 | Mean :1.6 | Mean :1.519 |
| 3rd Qu. | 3rd Qu.:2.000 | 3rd Qu.:2.000 | 3rd Qu.:2.000 | 3rd Qu.:3.000 | 3rd Qu.:2.000 | 3rd Qu.:3.000 | 3rd Qu.:2.0 | 3rd Qu.:2.000 |
| Max | Max. :4.000 | Max. :4.000 | Max. :4.000 | Max. :4.000 | Max. :4.000 | Max. :4.000 | Max. :4.0 | Max. :4.000 |
| NA’s | NA’s :25 | NA’s :27 | NA’s :15 | NA’s :48 | NA’s :28 | NA’s :62 | NA’s :33 | NA’s :40 |
DataIT_clean <- na.omit(DataIT[, c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num", "fltlnl_num", "enjlf_num", "fltsd_num", "cldgng_num")])- However, not to undermine the statistical power, is better to replace missing values with the column mean to preserve sample size.
for (col in c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num","fltlnl_num", "enjlf_num", "fltsd_num", "cldgng_num")) {
DataIT[is.na(DataIT[, col]), col] <- mean(DataIT[, col], na.rm = TRUE)}- Now the Cronbach’s Alpha can be computed.
suppressPackageStartupMessages(library(ltm))
cronbach_alpha <- cronbach.alpha(DataIT[, c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num","fltlnl_num", "enjlf_num","fltsd_num", "cldgng_num")], na.rm = TRUE)
cronbach_alpha##
## Cronbach's alpha for the 'DataIT[, c("fltdpr_num", "flteeff_num", "slprl_num", "wrhpp_num", ' ' "fltlnl_num", "enjlf_num", "fltsd_num", "cldgng_num")]' data-set
##
## Items: 8
## Sample units: 2865
## alpha: 0.838cat("Cronbach’s alpha =", round(cronbach_alpha$alpha, 3))## Cronbach’s alpha = 0.838- Given the independent variables mentioned before, namely the perceived life control (ctrlife), life satisfaction (stflife), trust in people (ppltrst) and satisfaction related to climate change action (testji9) and health sevices (stfhlth), I am going to compute the correlation between independent variables and the dependent-independent one.
DataIT$ctrlife <- as.numeric(as.character(DataIT$ctrlife))
DataIT$stflife <- as.numeric(as.character(DataIT$stflife))
DataIT$ppltrst <- as.numeric(as.character(DataIT$ppltrst))
DataIT$testji9 <- as.numeric(as.character(DataIT$testji9))
DataIT$stfhlth <- as.numeric(as.character(DataIT$stfhlth))
subset <- DataIT[, c("ctrlife", "stflife", "ppltrst", "testji9", "stfhlth", "CES_D8")]
correlation <- cor(subset, use = "complete.obs")
library(knitr)
kable(round(correlation, 2), caption = "Correlation Matrix of Key Variables")| ctrlife | stflife | ppltrst | testji9 | stfhlth | CES_D8 | |
|---|---|---|---|---|---|---|
| ctrlife | 1.00 | 0.45 | 0.21 | 0.22 | 0.12 | -0.38 |
| stflife | 0.45 | 1.00 | 0.23 | 0.08 | 0.19 | -0.45 |
| ppltrst | 0.21 | 0.23 | 1.00 | 0.05 | 0.16 | -0.17 |
| testji9 | 0.22 | 0.08 | 0.05 | 1.00 | 0.04 | -0.13 |
| stfhlth | 0.12 | 0.19 | 0.16 | 0.04 | 1.00 | -0.11 |
| CES_D8 | -0.38 | -0.45 | -0.17 | -0.13 | -0.11 | 1.00 |
This correlation matrix gives a first hint about how variables interact and allows to explore the patters and relationships between variables before proceeding with the multivariate regression analysis.
4. INITIAL RESULTS
In order to understand if the chosen independent variable have a
positive or negative influence on depressive symptoms, a multivariate
regression model using the lm function will be developed.
This
aims to understand how the independent variables I chose
actually impact the CES-D8 Scale. Given the literature review
and the correlation matrix presented above, I expect that when my
independent variables increase by 1 unit, depression decreases.
In other words, I expect that a higher perceived control and
satisfaction over one’s life, increased trust feelings toward others, as
well as a better planned healthcare services and government action in
the climate field help reducing depressive symptoms.
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 16.794 | 0.819 | 20.515 | 0.000 |
| ctrlife | -0.521 | 0.104 | -4.994 | 0.000 |
| stflife | -0.877 | 0.101 | -8.713 | 0.000 |
| ppltrst | -0.086 | 0.077 | -1.113 | 0.266 |
| testji9 | -0.111 | 0.069 | -1.602 | 0.110 |
| stfhlth | -0.027 | 0.081 | -0.340 | 0.734 |
The regression model explains approximately 24.89% of the
variance in depressive symptoms (CES_D8), indicating a moderate
explanatory power (Adjusted R² = 0.2428). The model is
statistically significant overall (F = 40.76, p <
1.195737e-71).
Among the predictors, perceived control over
life (ctrlife) and life satisfaction (stflife) are the strongest and
statistically significant factors negatively associated with
depressive symptoms. Specifically, greater control reduces distress by
-0.521 units, and higher life satisfaction reduces it by -0.877
units.
However, we might need to weight the regression model. With a Post-Stratification weight we ensure sampling and non-response errors are minimized.
| Estimate | Std. Error | t value | Pr(>|t|) | |
|---|---|---|---|---|
| (Intercept) | 16.848 | 0.819 | 20.577 | 0.000 |
| ctrlife | -0.508 | 0.103 | -4.934 | 0.000 |
| stflife | -0.878 | 0.100 | -8.809 | 0.000 |
| ppltrst | -0.070 | 0.078 | -0.893 | 0.372 |
| testji9 | -0.161 | 0.069 | -2.344 | 0.019 |
| stfhlth | -0.002 | 0.081 | -0.030 | 0.976 |
Through the Anova Model we can now compare the
weighted and unweighted regression models and check how
the adjusted sample address representativeness.
anova_results <- anova(model1, model2)
anova_results %>%
kable(caption = "ANOVA Comparison of Two Models", digits = 3)| Res.Df | RSS | Df | Sum of Sq | F | Pr(>F) |
|---|---|---|---|---|---|
| 615 | 7692.395 | NA | NA | NA | NA |
| 615 | 7461.355 | 0 | 231.04 | NA | NA |
As the Anova model shows, the Residual Sum of
Squares of the weighted model is slightly smaller than the unweighted
model’s one. This usually indicates that the weighted regression
analysis improves the model’s fit, but in this case, as the
difference is small, it might not be statistically significant either.
All in all, among the predictors, perceived control over life
(ctrlife) and life satisfaction (stflife) are the strongest and
statistically significant factors negatively associated with
depressive symptoms. Specifically, greater control reduces distress by
0.521 units, and higher life satisfaction reduces it by 0.877
units.
Other variables—trust in people, confidence in government climate action, and health satisfaction—do not show statistically significant effects, indicating weaker associations with depressive symptoms.
5. PREDICTORS AND OUTCOME OF CLINICALLY SIGNIFICANT DEPRESSION
In order to reduce the limitations of this study and provide an
objective parameter to evaluate the depressive symptoms, I am going to
set a cut off rule to reveal how many Italian people in this sample can
be considered clinically depressed.
Given, the CES-D8 Scale ranges
between 0 and 24, Briggs R. et al (2018) found that the reasonable cut
off threshold for such a Scale should be >= 9.
In order to do that, I’ll create binary variable for clinical depression (1 = yes, 0 = no)
DataIT$depression<- ifelse(DataIT$CES_D8 >= 9, 1, 0)
table(DataIT$depression)##
## 0 1
## 2115 650# and now I can calculate the final proportion of individuals above vs below the cutoff
prop.table(table(DataIT$depression))##
## 0 1
## 0.7649186 0.2350814Now I, given the binary outcome, I can create a logistic regression model GLM
modelGLM <- glm(DataIT$depression ~ ctrlife + stflife + ppltrst + testji9 + stfhlth,
data = subset, family = binomial(link = "logit") )
summary(modelGLM)##
## Call:
## glm(formula = DataIT$depression ~ ctrlife + stflife + ppltrst +
## testji9 + stfhlth, family = binomial(link = "logit"), data = subset)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 4.14757 0.60688 6.834 8.24e-12 ***
## ctrlife -0.33721 0.07200 -4.684 2.82e-06 ***
## stflife -0.45566 0.06948 -6.558 5.45e-11 ***
## ppltrst -0.02526 0.05574 -0.453 0.650
## testji9 -0.05728 0.05053 -1.134 0.257
## stfhlth 0.05698 0.05842 0.975 0.329
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 708.66 on 620 degrees of freedom
## Residual deviance: 584.82 on 615 degrees of freedom
## (2244 observations deleted due to missingness)
## AIC: 596.82
##
## Number of Fisher Scoring iterations: 4exp(coef(modelGLM))## (Intercept) ctrlife stflife ppltrst testji9 stfhlth
## 63.2801140 0.7137591 0.6340314 0.9750583 0.9443309 1.0586325exp(confint(modelGLM))## 2.5 % 97.5 %
## (Intercept) 19.8036574 214.5665056
## ctrlife 0.6182933 0.8204693
## stflife 0.5518080 0.7249731
## ppltrst 0.8741304 1.0880254
## testji9 0.8551529 1.0428418
## stfhlth 0.9447589 1.1883775r_mcfadden = with(summary(modelGLM), 1 - deviance/null.deviance)
r_nagelkerke = with(summary(modelGLM), r_mcfadden/(1 - (null.deviance / nrow(modelGLM$data)*log(2))))
r_nagelkerke## [1] 0.2109136The logistic regression model shows that higher levels of perceived
control over life (ctrlife) and life satisfaction (stflife) are
significantly associated with lower odds of the outcome, with odds
ratios of 0.71 and 0.63, respectively, and 95% confidence intervals that
do not include 1 (ctrlife: 0.62–0.82; stflife: 0.55–0.72).
Other
variables such as trust in people (ppltrst), trust in the justice system
(testji9), and self-rated health (stfhlth) have odds ratios close to 1,
suggesting weaker or non-significant effects.
The model explains
approximately 21.1% of the variance in the outcome, as indicated by
Nagelkerke’s R² = 0.211.
6. CONCLUSIONS
This study confirms that perceived life control and life
satisfaction are the most significant predictors of lower depressive
symptoms among Italians, reinforcing previous findings by
Nguyen et al. (2020) and Zalewska et al. (2021).
In contrast, trust in people, healthcare services, and government
climate action did not show significant associations with psychological
distress, diverging from earlier research (e.g., Estrada et al., 2019;
Ahnquist et al., 2010; Shen et al., 2024).
These results suggest that while institutional trust may have
indirect effects, subjective well-being and personal agency are more
impactful for mental health in this context.
The final step of this paper consists of defining a threshold to identify how many people in the sample can be actually defined as “clinically depressed” according to Briggs R. et al. (2018). About 681 people out of 2.865, namely the 24% of the sample, can be considered clinically depressed and the variance in the logistic regression model is about the 21.1%, which is credible.
However, the study has limitations: its
cross-sectional design restricts causal claims, reliance on
self-reported data introduces potential bias, and the model explains
only approximately 24.28% of the variance. Possible multicollinearity
between life satisfaction and life control may have also affected
estimates.
Future research should explore the pathways linking institutional trust and mental health, consider cultural and regional differences, and evaluate interventions aimed at boosting life satisfaction and perceived control.
6. REFERENCES
Ahnquist, J., Wamala, S. P., & Lindström, M. (2010). What has trust in the health-care system got to do with psychological distress? Analyses from the national Swedish survey of public health. International journal for quality in health care : journal of the International Society for Quality in Health Care, 22(4), 250–258. https://doi.org/10.1093/intqhc/mzq024
Martínez, L. M., Estrada, D., & Prada, S. I. (2019). Mental health, interpersonal trust and subjective well-being in a high violence context. SSM - population health, 8, 100423. https://doi.org/10.1016/j.ssmph.2019.100423
Nguyen, T.- vy, McPhetres, J., & Deci, E. L. (2020). Beyond God and Government: The Role of Personal Control in Supporting Citizens’ Well-Being in the Face of Changing Economy and Rising Inequality. Social Psychological Bulletin, 15(1), 1-21. https://doi.org/10.32872/spb.2663
Rasanathan, K. (2024). How can health systems under stress achieve universal health coverage and health equity? International Journal for Equity in Health, 23(1). https://doi.org/10.1186/s12939-024-02293-2
Round 11 questionnaire and provisional release dates | European Social Survey. (2025, January 6). https://www.europeansocialsurvey.org/news/article/round-11-questionnaire-and-provisional-release-dates
Shen, S. (2024). Green Finance and Health: How Does Implementing Carbon Emissions Trading Affect Mental Health? Advances in Economics, Management and Political Sciences, 44(1), 253–261. https://doi.org/10.54254/2754-1169/44/20232191
Van Damme-Ostapowicz, K., Cybulski, M., Galczyk, M., Krajewska-Kulak, E., Sobolewski, M., & Zalewska, A. (2021). Life satisfaction and depressive symptoms of mentally active older adults in Poland: a cross-sectional study. BMC Geriatrics, 21(1). https://doi.org/10.1186/s12877-021-02405-5
Wu, Y., Wang, L., Tao, M., Cao, H., Yuan, H., Ye, M., Chen, X., Wang, K., & Zhu, C. (2023). Changing trends in the global burden of mental disorders from 1990 to 2019 and predicted levels in 25 years. Epidemiology and psychiatric sciences, 32, e63. https://doi.org/10.1017/S2045796023000756
Zhang, Y. (2024). The road home: intimacy with parents, trust, and depression. Humanities & Social Sciences Communications, 11(1). https://doi.org/10.1057/s41599-024-03433-3