Research Question
Is there a correlation between the percentage of ‘government expenditure on education, total (% of the government expenditure)’ and ‘Suicide Rate / 100k people’ worldwide?
1 Motivation and Importantance of the Question
1.1 Motivation for Choosing the Question
The percentage of ‘government expenditure on education, total (% of the government expenditure)’ is calculated by dividing the total government expenditure on education by the total government expenditure on all sectors. This measure indicates the government’s spending devoted toward education, allowing an assessment of how high of a priority education may be, relative to other public sectors. Additionally, with awareness towards mental health being highly pushed by media, with the stigma surrounding mental illness progressively reducing, we are intrigued to investigate both measures.
Surprisingly, with the stigma continuing to reduce particularly in young males, suicide rates continue to climb. Our group members have witnessed the impact of menta health and suicides amongst their respective community and do believe that governments around the world can limit this by assisting in education regardless of primary, secondary or tertiary.
Intrigued by this, we were curious to determine whether an increase in the percentage of government expenditure on education coincides with a reduction in the suicide rate and thus has a negative correlation. We believe that policy makers have a strong role in ensuring that education is deemed highly important for younger people and whether or not capital-towed education is being used effectively.
1.2 Importance of the Question and Topic Forming Basis of Hypothesis
We feel that the question we have decided to explore has a highly
important nature to it. Examining whether an increase in government
expenditure towards education correlates with a reduction in suicide
rates can deliver significant conclusions to government officials, which
could ultimately save lives.
A reduction in the number of young people suffering from mental
illnesses will improve the effectiveness of businesses and ultimately
the economy. Therefore, we a struggling to understand why the government
is not spending more money on ensuring a suitable and strong education
for young people. Our question may assist us in understanding what
countries have been successful in directing capital to education, which
can assist in obtaining employment, and result in having a sufficient
income to survive. It is currently estimated that 726,000 people
worldwide commit suicide, and throughout Australia it is the leading
cause of death amongst people age 15-24, further indicating the
importance of this matter. [^FOOTNOTE1] While it may seem simple that
increasing government expenditure directed towards education may
coincide with lower suicide rates, some research tends to disagree. That
they can attribute lower suicide rates with over aspects of government
expenditure, specifically infrastructure which produces greater job
opportunities. However, research our chosen question may assist us in
understanding whether increased expenditure on education is worthwhile,
or the government should continue to inject capital in other areas to
lower the suicide rate. Additionally, we must assess whether countries
and governments worldwide are firstly deploying resources to counter
growing suicide rates, or whether policy makers are rethinking their
approach to combat this issue. Governments have a large responsibility,
as everyday people are deciding to take their lives.
2 Hypothesis
2.1 Hypothesis
Is there a correlation between the percentage of ‘government expenditure on education, total (% of the government expenditure)’ and ‘Suicide Rate / 100k people’ worldwide?
2.2 The Original Data
As viewed in the above table, the original Gapminder dataset comprises of 1704 rows and 6 columns, consisting of 6 differing variables. These include: 1. Country 2. Continent 3. Year 4. lifeExp (life expectancy) 5. pop (population) 6. gdpPercap (GDP per capita)
3.1 Downloading the data
We accessed our new Gapminder data through using data directly from the official Gapminder website (https://www.gapminder.org/data/). Based on our hypothesis we decided that the following datasets would be best appropriate: 1. Government expenditure on education, total (% of the government expenditure) 2. Suicide Rate / 100k people’ worldwide
After reviewing the data in R Studio, we discovered some apparent issues throughout the data. Firstly, the data offered for the suicide rate per 100,000 people may have appeared outdated, as the data began in 1950 – 2016. Greater analysis indicated several missing data points with many nations.
Secondly, the data provided for Govt. expenditure on education starts recording in 1980, where was the suicide data is provided from 1950 onwards. This discrepancy of 30 years may make the longitudinal analysis difficult to perform, however if we consider data onwards from 1980, approximately 40 years may be long enough to show indications of correlation.
As a group we decided that these indicators listed would have been adequate to test our hypothesis, however we would’ve liked something more extensive.
We decided to view a more updated version of the data we had acquired beforehand, while ensuring it was delivered from a non-bias reputable source similar to Gapminder. ‘Our World in Data’, an organisation based in the UK, instead of using the previous indicators found in Gapminder, we decided to use the equivalent datasets acquired from Our World in Data, as all data started from at least 1980 onwards and went up too 2021. These datasets were: 1. Government expenditure on education, total (% of government expenditure) 2. Suicide death rate, per 100,000 people
## Rows: 3822 Columns: 4
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr (2): Entity, Code
## dbl (2): Year, Government expenditure on education, total (% of government e...
##
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.## Rows: 6840 Columns: 4
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr (2): Entity, Code
## dbl (2): Year, Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate)
##
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.## # A tibble: 6 × 4
## Entity Code Year Government expenditure on education, total (% of gov…¹
## <chr> <chr> <dbl> <dbl>
## 1 Afghanistan AFG 2006 14.6
## 2 Afghanistan AFG 2007 10.8
## 3 Afghanistan AFG 2008 4.86
## 4 Afghanistan AFG 2009 6.49
## 5 Afghanistan AFG 2010 6.40
## 6 Afghanistan AFG 2011 5.15
## # ℹ abbreviated name:
## # ¹`Government expenditure on education, total (% of government expenditure)`## spc_tbl_ [3,822 × 4] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
## $ Entity : chr [1:3822] "Afghanistan" "Afghanistan" "Afghanistan" "Afghanistan" ...
## $ Code : chr [1:3822] "AFG" "AFG" "AFG" "AFG" ...
## $ Year : num [1:3822] 2006 2007 2008 2009 2010 ...
## $ Government expenditure on education, total (% of government expenditure): num [1:3822] 14.56 10.81 4.86 6.49 6.4 ...
## - attr(*, "spec")=
## .. cols(
## .. Entity = col_character(),
## .. Code = col_character(),
## .. Year = col_double(),
## .. `Government expenditure on education, total (% of government expenditure)` = col_double()
## .. )
## - attr(*, "problems")=<externalptr>## Entity Code Year
## Length:3822 Length:3822 Min. :1972
## Class :character Class :character 1st Qu.:2003
## Mode :character Mode :character Median :2010
## Mean :2009
## 3rd Qu.:2016
## Max. :2022
## Government expenditure on education, total (% of government expenditure)
## Min. : 0.00
## 1st Qu.:11.06
## Median :14.09
## Mean :14.42
## 3rd Qu.:17.14
## Max. :44.80The dataset represented above is ‘Government expenditure on education, total (% of government expenditure)’, including 3822 rows and 4 columns. The 4 variables in each of these columns are: 1. ‘Entity’ – country 2. ‘Code’ – country code 3. Year 4. Government expenditure on education, total (% of government expenditure)
## # A tibble: 6 × 4
## Entity Code Year Deaths - Self-harm - Sex: Both - Age: Age-standardiz…¹
## <chr> <chr> <dbl> <dbl>
## 1 Afghanistan AFG 1990 8.28
## 2 Afghanistan AFG 1991 8.17
## 3 Afghanistan AFG 1992 8.14
## 4 Afghanistan AFG 1993 8.23
## 5 Afghanistan AFG 1994 8.36
## 6 Afghanistan AFG 1995 8.41
## # ℹ abbreviated name:
## # ¹`Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate)`## spc_tbl_ [6,840 × 4] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
## $ Entity : chr [1:6840] "Afghanistan" "Afghanistan" "Afghanistan" "Afghanistan" ...
## $ Code : chr [1:6840] "AFG" "AFG" "AFG" "AFG" ...
## $ Year : num [1:6840] 1990 1991 1992 1993 1994 ...
## $ Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate): num [1:6840] 8.28 8.17 8.14 8.23 8.36 8.41 8.46 8.53 8.58 8.7 ...
## - attr(*, "spec")=
## .. cols(
## .. Entity = col_character(),
## .. Code = col_character(),
## .. Year = col_double(),
## .. `Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate)` = col_double()
## .. )
## - attr(*, "problems")=<externalptr>## Entity Code Year
## Length:6840 Length:6840 Min. :1990
## Class :character Class :character 1st Qu.:1997
## Mode :character Mode :character Median :2004
## Mean :2004
## 3rd Qu.:2012
## Max. :2019
## Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate)
## Min. : 1.630
## 1st Qu.: 6.657
## Median :10.500
## Mean :12.064
## 3rd Qu.:14.920
## Max. :90.060The dataset represented above is ‘Suicide death rate, per 100,000 people’. This dataset similarly shares 4 columns, however has 6840 rows. The 4 variables include: 1. ‘Entity’ – country 2. ‘Code’ – country code 3. Year 4. Deaths – Self Harm – Sex: Both – Age: Age standardised (rate, per 100k people)
3.2 Renaming the labels of Education Data and Suicide Data
##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unioneducation_data <- education_data %>%
rename(GovSpendingEducation = `Government expenditure on education, total (% of government expenditure)`)## # A tibble: 4 × 4
## Entity Code Year GovSpendingEducation
## <chr> <chr> <dbl> <dbl>
## 1 Afghanistan AFG 2006 14.6
## 2 Afghanistan AFG 2007 10.8
## 3 Afghanistan AFG 2008 4.86
## 4 Afghanistan AFG 2009 6.49#Suicide Data
suicide_data <- suicide_data %>%
rename(SuicideRate100k = 'Deaths - Self-harm - Sex: Both - Age: Age-standardized (Rate)')## # A tibble: 4 × 4
## Entity Code Year SuicideRate100k
## <chr> <chr> <dbl> <dbl>
## 1 Afghanistan AFG 1990 8.28
## 2 Afghanistan AFG 1991 8.17
## 3 Afghanistan AFG 1992 8.14
## 4 Afghanistan AFG 1993 8.23As stated previously with our data being presented in long format, we could use the rename function to specify what was in each column more clearly. We renamed ‘Government expenditure on education, total (% of government expenditure)’ as ‘GovSpendingEducation’. Similarly, for the suicide data, we renamed ‘Deaths – Self Harm – Sex: Both – Age: Age-standardised (rate, per 100k people)’ as ‘SuicideRate100k’. Having decided to do this for both datasets meant that avoiding lengthy names were possible to make the data have a greater presentability to the viewer.
4 Cleaning the data
4.1 Cleaning the GovSpending Data
education_data_year_avg <- education_data %>%
group_by(Year) %>%
summarise(average = mean(GovSpendingEducation))
head(education_data_year_avg, n = 4)## # A tibble: 4 × 2
## Year average
## <dbl> <dbl>
## 1 1972 17.5
## 2 1973 17.8
## 3 1974 17.6
## 4 1975 17.4Cleaning all data is vitally important when dealing with your main dataset. Constraining data to two variables allows for easier data visualisation. Irrelevant variables that don’t provide beneficial information to our question were to be removed, including column 1 ‘entity’ and column 2 ‘code’. The remaining two variables were column 3 ‘year’ and column 4 ‘GovSpendingEducation’.
Once the above process was complete, we employed the group_by function. The earliest year data was recorded, 1972 onwards to 2022. This allowed the variables for their respective years to be grouped together, where operations can be performed as their year. This allowed us to summarise the average Government Spending on Education per year on a global scale. The previous steps allowed us to produce a dataset revealing the global government spending on education per year.
4.2 Cleaning the Suicide Data
suicide_data_year_avg <- suicide_data %>%
group_by(Year) %>%
summarise(average = mean(SuicideRate100k))
head(suicide_data_year_avg, n = 4)## # A tibble: 4 × 2
## Year average
## <dbl> <dbl>
## 1 1990 13.0
## 2 1991 13.0
## 3 1992 13.1
## 4 1993 13.3Similar to the above, the suicide dataset required cleaning. First, column 1 ‘entity’ and column 2 ‘code’ were removed from the dataset, leaving the remaining 2 ‘year’ and ‘SuicideRate100k’.
Furthermore, we established the group_by function by ‘year’ and entered it into the summarise function. By entering the dataset into the average function, the average suicide rate per year ranging from 1990 onwards was produced.
4.3 Merging of the Government Education Spending and Suicide data clean
MergedDataset <- inner_join(education_data_year_avg, suicide_data_year_avg, by = "Year")
MergedDataset <- MergedDataset %>%
rename(GovEducationSpending = average.x, AverageSuicideRate = average.y)
head(MergedDataset, n = 4)## # A tibble: 4 × 3
## Year GovEducationSpending AverageSuicideRate
## <dbl> <dbl> <dbl>
## 1 1990 13.9 13.0
## 2 1991 13.8 13.0
## 3 1992 14.5 13.1
## 4 1993 14.7 13.3The data cleaning of the government spending and suicide data was essential to ensure the merging of datasets was successful. This permits observation and analysis if a correlation may be present between the two variables. The inner_join function was used to merge all education spending as a percentage of total government expenditure and the death rate from suicides (per 100,000 people) together by the parallel years.
If data from either set did not have matching yearly values, they were removed, resulting in a new dataset from 1990 – 2019 with three complete variables. These three being: 1. Year 2. GovEducationSpending 3. AverageSuicideRate
We decided to name the dataset MergedDataset to make it understandable and visualise the data clearly.
Finally, the data is ultimately clear for interpretation and will be enabled to be coded into visualisation models, to examine whether there is a correlation between the percentage of ‘government expenditure on education, total (% of the government expenditure)’ and ‘Suicide Rate / 100k people’ worldwide.
4.4 Graphing the Dataset
ggplot(MergedDataset, aes(GovEducationSpending, AverageSuicideRate)) +
geom_point() +
geom_smooth(method=lm, se=F, col="red", linetype="dashed") +
labs(title = "The Effect of Increased Government Education Spending on Average Suicide Rates (Global)",
x = "Government Education Spending (% of Total Expenditure)",
y = "Average Suicide Rates (Global)")## `geom_smooth()` using formula = 'y ~ x'
model <- lm(AverageSuicideRate ~ GovEducationSpending, data = MergedDataset)
r_squared <- summary(model)$r.squared
print(r_squared)## [1] 3.628827e-05# Remove the year 1994 from the dataset
No94Dataset <- subset(MergedDataset, Year != 1994)
# Plot with the filtered data
ggplot(No94Dataset, aes(GovEducationSpending, AverageSuicideRate)) +
geom_point() +
geom_smooth(method=lm, se=F, col="red", linetype="dashed") +
labs(title = "Effect of Gov Education Spending on Suicide Rates (Excluding 1994)",
x = "Government Education Spending (% of Total Expenditure)",
y = "Average Suicide Rates")## `geom_smooth()` using formula = 'y ~ x'
No94model <- lm(AverageSuicideRate ~ GovEducationSpending, data = No94Dataset)
r_squaredNo94 <- summary(No94model)$r.squared
print(r_squaredNo94)## [1] 0.0266417We had decided that the greatest way to demonstrate any form of linear correlation, or relationship between the two selected variables would be through a scatter plot. To access this feature we were required to download the ‘ggplot2’ package. In the formulation of creating a scatter plot we had decided that our independent variable (x) would be “Government Education Spending” and our dependent variable (y) would be “Average Suicide Rates”. This should allow us to measure the effect spending has on the suicide rates.
The ‘MergedDataset’ created previously was used as the main dataset which held both X and Y variables under the aesthetics (aes) input. Additionally the geom_point function created a scatter plot of the points for where each year’s data stands in relation with one another. The geom_smooth function tells the program to place a linear regression line or line of best fit throughout the data to demonstrate the relationship between both X and Y. For suitable visualisation of the graph the aes() function and labs() function was used to implement correct axis and graph titles, as well as format the inputs correctly.
We finally, implemented the lm() function which allowed for a linear model of our data to further find the R-squared value which represents coefficient of determination value showing how well the data fits the model.
As viewed in the graph there is a lack of a negative correlation between government education spending and average suicide rate. As displayed by the r-squared result this reinforces the lack of correlation (), where this value is near zero. This result not supporting our hypothesis.
The graph shows a varying degree of scatter surrounding the negative correlation as some data begins in the top left and trends downward. However, in 1994 the government spending peaked at 16.5% (% of total expenditure), but average suicide rates ranked second during this year as well. After removing the 1994 outlier, the new coefficient of determination resulted in 0.0266, which still remained inconclusive regarding our hypothesis, however represented a stronger visualisation against the graph displaying a negative correlation may be present.
4.5 Distribution of Merged Dataset
ggplot(MergedDataset, aes(x = GovEducationSpending)) + geom_histogram(binwidth = 0.20, col = "white") + labs(title = "Normal Distribution of % Government Education Spending", x = "Government Education Spending (% of Total Expenditure)", y = "Frequency")
Data distribution is a highly relevant function within statistical analysis, representing all values, while calculating the relative frequency. Through R-Studio it allowed analysis of the distribution of ‘Government Education Spending’, which was indicative of the frequency of the differences in government spending. Through the normal distribution representing a bell curve shape, this distribution shows the data near the mean are more frequent in occurrence. Hence, in a visual form the normal distribution can be viewed as a “bell curve”
To graph the ‘MergedDataset’ distribution, the GGPLOT2 function has to be installed, where the geom_histogram() function specifically sets it to the histogram set up. Additionally, the labs() function was used to label the histogram with appropriate labels on either axis. Through deciding that the binwidth of 0.20 would represent the data in the normal distribution, it was apparent that the data was positively skewed. Although, majority of the data occurs around the mean of 14.49% as reflected by the bell curve shape.
Although a slight bell curve shape is apparent, a weakness shown is the lack of data to display an entire bell curve. We can assume that having more datapoints (more years), the data is likely to show a greater pattern visually displaying a more symmetrical graph. Nevertheless, the data can be analysed to represent an indication of a bell curve, however having a singular outlier past the 16% mark. Given the positively skewed nature of the histogram, the mean is slightly to the right of the highest point, however the median and mode would be left of the mean, resulting in the skewed shape.
5 Regression Analysis
5.1 Install Packages
The “infer” package allows for statistical inference. This gives the ability to create a hypothesis which can be specified and generated, these statistics can be created and calculated.
RegressionAnalysis <- lm(AverageSuicideRate ~ GovEducationSpending, data = MergedDataset)
summary(RegressionAnalysis)##
## Call:
## lm(formula = AverageSuicideRate ~ GovEducationSpending, data = MergedDataset)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.8665 -1.0248 0.2027 1.0274 1.4086
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 12.201178 4.299260 2.838 0.00835 **
## GovEducationSpending -0.009442 0.296207 -0.032 0.97480
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.127 on 28 degrees of freedom
## Multiple R-squared: 3.629e-05, Adjusted R-squared: -0.03568
## F-statistic: 0.001016 on 1 and 28 DF, p-value: 0.9748## [1] 0.7062175## [1] 1.106931## [1] 14.49776## [1] 12.06429## Year GovEducationSpending AverageSuicideRate
## Min. :1990 Min. :13.12 Min. :10.20
## 1st Qu.:1997 1st Qu.:14.17 1st Qu.:11.04
## Median :2004 Median :14.51 Median :12.27
## Mean :2004 Mean :14.50 Mean :12.06
## 3rd Qu.:2012 3rd Qu.:14.75 3rd Qu.:13.09
## Max. :2019 Max. :16.51 Max. :13.48t.test(MergedDataset$GovEducationSpending, MergedDataset$AverageSuicideRate, alternative = "two.sided", mu = 0, paired = T)##
## Paired t-test
##
## data: MergedDataset$GovEducationSpending and MergedDataset$AverageSuicideRate
## t = 10.124, df = 29, p-value = 4.992e-11
## alternative hypothesis: true mean difference is not equal to 0
## 95 percent confidence interval:
## 1.941846 2.925105
## sample estimates:
## mean difference
## 2.433475In terms of the rationale and methodology the process used to determine the relationship between government expenditure on education and suicide rates, a linear regression was utilised. This analysis will give clarity as to how expenditure causes variation in different countries when looking at suicide rates.
The dependant variable is clearly defined as the “suicide rate per 100,000 people.” Whilst the independent variable can be determined as the government expenditure on education when measured against the percentage of the total expenditure outputted by the government.
The linear regression analysis must be used to show the correlation between the p-value and the R-squared value to truly determine significance.
Functions lm(), mean(), sd() and summary() were all utilized in order to present the data in a fashion which is applicable.
5.2 Data Analysis
The regression analysis showed that Ordinary Least Squares Regression (OLS) The value of the R-squared model proves to be -0.03568, indicating that there is less than 0.4% variation in suicide rates differentiating countrites through government spending, highlighting a very weak relationship between variables.
The education expenditure coefficient value is determined as
-0.009442, highlighting a value of 0.9748 for “p.” Thus, highlighting
that both variables do not have a strong effect on one another.
Lastly the hypothesis testing is crucial to determine connection. The
null hypothesis shows that there is no relationship between government
expenditure and suicide rates b1=0
Alternative hypothesis shows there is a relationship between variables when b/=/0
The t-test is used to show a p-value of 4.9926e-11. This shows that this value is significantly below 0.05.
5.3 Evaluation
In analysing the data generated from comparing government expenditure on education to suicide rates there is a clearly established relationship, however the correlation between variables is weak. The R-squared value determines the strength, and a value of -0.03568 shows that the correlation cannot predict to a high determinant if govt expenditure implicates suicide rates internationally.
References
Our World In Data. “Death Rate from Suicides.” Our World in Data, ourworldindata.org/grapher/suicide-death-rates. Accessed 20 Sept. 2024.
“Share of Education in Government Expenditure.” Our World in Data, ourworldindata.org/grapher/share-of-education-in-government-expenditure. Accessed 20 Sept. 2024.
World Health Organization. “Suicide.” World Health Organization, World Health Organization, 29 Aug. 2024, www.who.int/news-room/fact-sheets/detail/suicide. Accessed 22 Sept. 2024.
Worldometer. “Countries in the United Nations - Worldometer.” Worldometers.info, 2024, www.worldometers.info/united-nations/#note16. Accessed 30 Sept. 2024.