R Notebook

The Effect of Gender and Age on Suicides Among Population of Different Countries

I will use the Suicide Rates Overview 1985 to 2016 data from Kaggle that compares socio-economic information with suicide rates by year and country to explore the effect of gender and age on the suicide rates in these countries. This data has information of two conceptual levels, individual and country-based. Individuals with their demographic characteristics are nested within the countries. My hypothesis is that there is a correlation between sex and suicide rates. Men commit suicide at highter rates than women across all age groups.

I will work with the following packages: library(nlme), library(dplyr), library(magrittr), library(tidyr), library(haven), library(lmerTest), library(ggplot2), library(texreg).

Now, importing data for analysis:

library (readr)
master<-read_csv("C:/Users/Marcy/Documents/soc 712/master.csv")

Parsed with column specification:
cols(
  country = col_character(),
  year = col_double(),
  sex = col_character(),
  age = col_character(),
  suicides_no = col_double(),
  population = col_double(),
  `suicides/100k pop` = col_double(),
  `country-year` = col_character(),
  `HDI for year` = col_double(),
  `gdp_for_year ($)` = col_number(),
  `gdp_per_capita ($)` = col_double(),
  generation = col_character()
)

head (master)

First, I will ignore the countly level data and analyzis the data on individual level by performing complete pooling model.

cpooling <- lm(suicides_no ~ sex, data = master)
summary(cpooling)

Call:
lm(formula = suicides_no ~ sex, data = master)

Residuals:
    Min      1Q  Median      3Q     Max 
 -373.0  -325.0  -111.1   -57.1 21965.0 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)  112.114      7.568   14.81   <2e-16 ***
sexmale      260.920     10.703   24.38   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 892.6 on 27818 degrees of freedom
Multiple R-squared:  0.02092,   Adjusted R-squared:  0.02088 
F-statistic: 594.3 on 1 and 27818 DF,  p-value: < 2.2e-16

Based on complete pooling model shown above, sex is a sole considered factor while running this linear model. The coefficient of sex is statistically very significant. Evidently, males commit suicides at more than double rate than females, 261 to 112.

Now, I will run a no-pooling model to conduct an effect of sex on suicide rates within countries to see if there is any variance exists.

dcoef <- master %>% 
    group_by(`country`) %>% 
    do(mod = lm(suicides_no ~ sex, data =.))
coef <- dcoef %>% do(data.frame(intc = coef(.$mod)[1]))
ggplot(coef, aes(x = intc)) + geom_density()+xlab("country")

dcoef=master %>% 
    group_by(`country`) %>% 
    do(mod = lm(suicides_no ~ sex, data = .))
coef <- dcoef %>% do(data.frame(difference = coef(.$mod)[2]))
ggplot(coef, aes(x = difference)) + geom_histogram()+xlab("Difference in Female and Male Suicide Rates by Country")

*As shown above, on average, the vast majority of the countries have reported between 0 and 500 suicides in each given year. However, in a few of them, the number of suicides is significantly higher, reaching over 1000. As per difference in male and female variation by country, it is also ranging greatly.

Now, I will use a random effect model to allow for group variation within our regression model. Also, I add the interaction of sex and age as the effect of combination of age and sex, which perspectively could have a significant added effect on suicide rates.

randomeffect=lme(suicides_no ~ sex*age, data = master, random = ~1|country, method = "ML")
summary(randomeffect)

Linear mixed-effects model fit by maximum likelihood
 Data: master 

Random effects:
 Formula: ~1 | country
        (Intercept) Residual
StdDev:    522.4601 655.4487

Fixed effects: suicides_no ~ sex * age 

 Correlation: 
                       (Intr) sexmal a25-3y a35-5y a5-14y a55-7y ag75+y s:25-y s:35-y s:5-1y
sexmale                -0.178                                                               
age25-34 years         -0.178  0.500                                                        
age35-54 years         -0.178  0.500  0.500                                                 
age5-14 years          -0.178  0.499  0.499  0.499                                          
age55-74 years         -0.178  0.500  0.500  0.500  0.499                                   
age75+ years           -0.178  0.500  0.500  0.500  0.499  0.500                            
sexmale:age25-34 years  0.126 -0.707 -0.707 -0.354 -0.353 -0.354 -0.354                     
sexmale:age35-54 years  0.126 -0.707 -0.354 -0.707 -0.353 -0.354 -0.354  0.500              
sexmale:age5-14 years   0.126 -0.706 -0.353 -0.353 -0.707 -0.353 -0.353  0.499  0.499       
sexmale:age55-74 years  0.126 -0.707 -0.354 -0.354 -0.353 -0.707 -0.354  0.500  0.500  0.499
sexmale:age75+ years    0.126 -0.707 -0.354 -0.354 -0.353 -0.354 -0.707  0.500  0.500  0.499
                       s:55-y
sexmale                      
age25-34 years               
age35-54 years               
age5-14 years                
age55-74 years               
age75+ years                 
sexmale:age25-34 years       
sexmale:age35-54 years       
sexmale:age5-14 years        
sexmale:age55-74 years       
sexmale:age75+ years    0.500

Standardized Within-Group Residuals:
        Min          Q1         Med          Q3         Max 
-5.24358160 -0.18933215  0.05006066  0.21000191 27.50033170 

Number of Observations: 27820
Number of Groups: 101 

Thus, adding effect of age groups provides notable results. Sex, age, and interaction of sex and age have sifnificant effect of suicide rates of the population. The standart deviation is 522. Males of 35-54 years of age are the most vulnerable group (as much as three times higher) for attempting suicides. However, reaching 75+ years, their risk of commiting suicide drops significantly to -107 which is below many other age groups and even below women of 75+ age whose number is 20 in this group. Overall, the lowest rates of suicide are among the age group of children and teenagers. Interestingly, young age of males between 5-14 committ suicides at the rate of -189 compared to this particupar age group of females whose number is -69. Thus, I can conclude that children are less likely to commit a suicide than older generation people, but female children are more than twice more likely to attempt suicides than male children.

Now, I will use a random slope affect model. Unlike a random intercept model, a random slope model allows each group line to have a different slope and that means that the random slope model allows the explanatory variable to have a different effect for each group.

Slope=lme(suicides_no ~ sex*age, data = master, random = ~sex|country, method = "ML")
summary(Slope)

Linear mixed-effects model fit by maximum likelihood
 Data: master 

Random effects:
 Formula: ~sex | country
 Structure: General positive-definite, Log-Cholesky parametrization
            StdDev   Corr  
(Intercept) 227.4749 (Intr)
sexmale     608.3545 0.929 
Residual    564.0490       

Fixed effects: suicides_no ~ sex * age 

 Correlation: 
                       (Intr) sexmal a25-3y a35-5y a5-14y a55-7y ag75+y s:25-y s:35-y s:5-1y
sexmale                 0.706                                                               
age25-34 years         -0.324  0.132                                                        
age35-54 years         -0.324  0.132  0.500                                                 
age5-14 years          -0.323  0.131  0.499  0.499                                          
age55-74 years         -0.324  0.132  0.500  0.500  0.499                                   
age75+ years           -0.324  0.132  0.500  0.500  0.499  0.500                            
sexmale:age25-34 years  0.229 -0.186 -0.707 -0.354 -0.353 -0.354 -0.354                     
sexmale:age35-54 years  0.229 -0.186 -0.354 -0.707 -0.353 -0.354 -0.354  0.500              
sexmale:age5-14 years   0.228 -0.186 -0.353 -0.353 -0.707 -0.353 -0.353  0.499  0.499       
sexmale:age55-74 years  0.229 -0.186 -0.354 -0.354 -0.353 -0.707 -0.354  0.500  0.500  0.499
sexmale:age75+ years    0.229 -0.186 -0.354 -0.354 -0.353 -0.354 -0.707  0.500  0.500  0.499
                       s:55-y
sexmale                      
age25-34 years               
age35-54 years               
age5-14 years                
age55-74 years               
age75+ years                 
sexmale:age25-34 years       
sexmale:age35-54 years       
sexmale:age5-14 years        
sexmale:age55-74 years       
sexmale:age75+ years    0.500

Standardized Within-Group Residuals:
         Min           Q1          Med           Q3          Max 
-10.03880367  -0.16086250   0.01425447   0.16910246  27.96197319 

Number of Observations: 27820
Number of Groups: 101 

Data results reveal that there is a significant gender and age correlation between suicide rates of population in all reported countries. There is also significant variation in the number of suicide rates in the countires ranging from less then 10 to over 1000. Across the board, men are more than twice likely to commit a suicide. Adding age as an interaction term with person’s gender plays a very important role. Men of age 35-54 are in the higherst risk group among those who are likely to committ suicide. Being a male, your chances of committing a suicide are 150.8 against being a women with 56.7. If you also consider the age, the chances of males between 34-54 years who commit suicides are alarmingly tripled.

Now, I will compare these three models to see which one fits data best.

AIC(cpooling, randomeffect, Slope)

Apparently, Randon Slope Model fits the data best as its AIC value is less than of other two models.

To conclude on my hypothesis, it is true that men commit suicide at highter rates than women, but there is a signifciant gender variation across different age groups as shown in my tables.