Introduction

I take a look at the frequecy of how non-medical causes of death trend over a time span of 38 years in Germany, Russia, Philippines, United Kingdom (UK), United States(US), Brazil, Canada, Japan, and Mexico. I compared the trends to see how they differed in the various countries. Looking at these death cause rates can provide information which can potentailly be used to medigate (likely through preventative safety measures, early detection, or education) the frequency of death by external factors unrelated to an individual’s physical health. It is also interesting to consider deathrate trends in relationship to other external factors such as cultural norms and political climate.

Data

This dataset is derived from the World Health Organization (WHO) Mortality datatbase; source: http://apps.who.int/healthinfo/statistics/mortality/whodpms/ WHO offers this data set in two forms: the first as raw data where each death represents exactly one person, creating a 1:1 incidence ratio. The second is in age-standardized death rates per 100,000 world standard population, a weighted value which allows for comparison between countries. I make use of the latter for all the following computations.

Method

First I made line plots for each cause of death over time per country using the plot and line functions. I organized the countries by color where Russia is represented by magenta, Germany;red, Philippines;orange, UK;teal, US;blue, Brazil;yellow, Canada;indigo, Japan:purple, Mexico;green. From there I recognized that deathrates for different causes in the varying countries had vastly different correlation trends and strengths over time. To quantify this I calculated correlation statistics for each country per deathrate with Benjamini-Hochberg corrected p-values. Aferwards I calculated the effect size which is the sign of the correlation times(*) the -log base e of the corrected p-value. Using the effectsize of each mode of death I made bar plots in order to compare different countries by deathrates.

From the original line plots I noted that there was missing data which may reduce the power to detect significant correlation, thereby impacting the effect size. As a case study, I imputed the values of suicide rates in the countries with missing values; Germany, Phllilpines and Russia and demonstrated an increased power in the correlation analysis. Finally I created a set of parallel barplots of original effect size and imputed value effect size for a clear visual display.

Results

Overview I refer to the details of the line and bar plots in order from left to right, reading each of the four columns (c) from top row to bottom row (r). Each line and bar plot represents a different external cause of death and the slopes/ effect size for each country is displayed and represented by a different color.

Line Plots Looking at the line plots it is evident that there is one country, Russia, that consistantly has the highest death rates across all categories, less two: death by falls (fig1:c1,r2) and death by assault (fig1:c4,r1). Also Russia’s line plots also seem to be the most sparatic making it difficult to spot neither an acending or decending slope line. All the other countries’ trend lines can be clearly seen to move in one direction or another. Death as a result of falling is highest in the United States, represented by the blue line. Death by assault is highestin Brazil, the yellow line.

Bar Graphs Looking at the bar plots below, a negative effect size extends downwards away from the zero mark and and a positive effect size rises above zero. Therfore a large effect size indicates a low p-value for the slope and that the slope of that line is significantly different from zero. In other words the taller the bar, signifies a steeper slope and stronger correlation. For example, looking at death by transport, the trend in Germany is stronger than that of Brazil. Anything that falls within the significace thresholds(marked by dashed lines) is not significant in p-value. If the bar extends past the significance threshold in either direction, then it is significant; above the line indicates a positive correlation and below indicates a negetive one.

Death by Transport

The slopes for all countries has a negative correlation over time with the exception of the Philippines which has a positive correlation (fig1: c1,r1). According to the bar plot (fig2: c1,r1), with the exception of Russia, Philippines and Brazil, all the countries have a negative and tall bar indicating a stronger correlation.

Death by Falls

Some trends appear to slope downward and others upwards, the bar plot shows the same where Russia, Philippines, US and brazil have significantly positive effect sizes and the remaining counties have significantly negative effect sizes.

Death by Drowning

With the exception of Philippines, all countries show a significantly negative correlation in drowning over time with Mexico having the strongest correlation, meaning it also has the steepest slope. Philippines is the only country to have a positive slope, however it it within the significance thresholds wich means it is a positive slope with a high p-value.

Death by Smoke inhalation or Fire

Both Russia and Philippines have slightly significant p-values, with bar plots that just barely extend past the significance threshold with the effect size for Russia being just slightly larger that the Philippines. All other countries show a strong negative effect size that is significant in all cases, with the US and UK showing an equal level of strength.

Death by Poison

Although Russia has the highest rate of death by poison (fig1:c3,r1), the US has the strongest effect size (fig2:c3,r1). This shows us that althoguh a country’s death rate may be high, its trend might be consistant over time which means that time does not have an effect on this given variable. Therefore the effect size will not be significant and will most likely fall within the significance thresholds.

Death by Suicide

Germany, Russia, UK, and Canada all show significant effectsizes with a negative correlation over time, looking at the bar and line plots respectively. Canada has the strongest effect size at -20+. The US and Japan have non-significant effect sizes that fall within the significance thresholds. However note that Japan has the second highest suicide death rate compared to Russia with the hightest death rate (fig1: c3,r2).However, Russia is among those that has a negative trending slope with a low p-value and significant effect size.

Death by Assault

Brazil stands alone with a significantly positive effect size. Russia and the Philippines fall within the significance thresholds and all other countries have significantly negative trends over time.

Death by Other External Causes

Russia has the only line plot that is not decending. Looking at the bar plots we can see that all countries have significantly negative effect sizes with the exception of the UK and Japan whose bar plots do not hold a significant value in either direction.

Death Trends- Line Plots

Figure 1.

Effect size - Bar Graphs

Figure 2.

Case Study

A look at suicide rates with imputed values from the coutries with missing values: Phllilpines, Russia, and Germany

Note: The colored lines in the images above represent the average of 5 imputations, it is this line that is used to fill the gap of missing values from the original line plots.

Including Imputed Values

Discussion

After looking at the results for all the countries, I noticed that there was missing values for 3 different countries, Germany, Philippines and Russia. Coincidently each country also has a different case of the level of missing data. Philippines missing 54%, Russia missing 6%, and Germany missing 28%. Using the death by Suicide data set, I formed a case study to see if using imputed values would result in a more powerful effectsize and stronger correlation. I used the Amelia package to extrapulate outside the range of known values. Then used the average of five imputations for each of the three countries. The results showed that the 3 countries with missing data did show an even stronger correlation with the imputed values added. There are several different social or political implications that can be made from a careful examination of the results. Still using the suicide mode of death as an example, the significantly negative trend of Germany, UK, and Canada might be indicative of their respective government’s proactive approach to finding mental health solutions and free health coverage for their entire population. The US’s non-signifacant effect size , yet 3rd highest suicide rates might translate to the lack of mental health awareness or might be associated with high insurance costs. In another example, using the death by transportation, most countries showed a strong negative correlation over time. This is likey due to a worldwide standarized effort to increase car safety regulations, where both the car manufacturers and law enforcement work in tandum to increase the compliance. A further examination of each case would have to be done to discover all the possibilities of what social or political factors play a role in the varying modes of death.

Conclusion

Over time, external causes of death generaly show either a negative or positive correlation with varying stregths from country to country. With some death types like transport, drowning and poison, most countries follow a similar trend in the same direction but vary in correlation strength. The social implications for this might be because of standardized safety regulations surrounding these causes of death. However there are other modes of death where it seems the countries are split with polar trend lines of equal or varying strengths, as seen with suicides and death by falling, these differences might have to do with cultural views towards suicide or the amount of elderly population.

In the case of missing values, imputed values were used using the average of 5 imputations for the three countries that were missing the information: Germany, Russia, and Philippines. Using the imputed values showed an increased power in the correlation analysis observed with a side by side plot table.