Part A: Motivation of this project – why did we choose this topic?
The Economic Perspective of Suicide: It is interesting that we actually found this topic to be very controversial. There were some research in the 80’s and 90’s arguing that the economic loss for the whole society due to suicides (majorly due to the lost productivity) is significantly high. “There are about 30,000 individuals who completed suicide each year in the United States. According to an estimation of the costs to society of depression in 1980 calculated that the 16,111 depressed people who completed suicide in 1980 resulted in a loss of $4.2 billion due to lost productivity, which comes to a loss of $260,691 for each person who completed suicide.”
However, some other researchers (Yang, B., & Lester, D) redid the cost-benefit analysis a decade later, arguing that the total savings from people who suicide, in fact, exceeds the cost to the society, and therefore there is a net savings to the society as whole. Although it does not imply that suicide should be acceptable to society, but purely in economic sense, the prevention of suicide is likely to incur economic costs to the society as does the prevention and treatment of medical illness.
Why it is interesting and important: Obviously, researching on suicide is very important, no matter which of the two economic perspectives described is more reliable. If it is the case that the cost of suicide to the society is very significant that we need to do some prevention, then once we get the idea of what the important determinants are, we can focus on the ones have most impact. For example, if the higher the education level is, the fewer people tend to commit suicide, we can try to subsidize education and improve the overall education level of people, and then we expect suicide rate will be lower according to this action. If we find out that the suicide rate for some specific age group is relatively higher than the other groups, we may set up some program providing help specifically to this group. This research will also be interesting, because we want to examine if the regression result is in compliance with our common sense (what effect on suicide rate there will be for each different variable). This will be discussed in more details in part B.
Part B: Research design – motivation and estimation of independent / dependent variables
Dependent Variable (Y) – Suicide rate This is just what we want to do research on, and the way it is calculated is as following: Independent Variable (X)
Education impact (High school completion rate, and % of population get Bachelor’s degree or more), calculated as

People with higher education tend to be better at adapting themselves to shocks, difficulties and failures. So, we decide to use high school completion rate and % of people who get Bachelor’s degree or more as our regressor, and we expect the higher these two rates are, the lower the suicide rate should be (the coefficient should be negative).
Age impact (% of people aged between 15 and 44), calculated as: According to the research of Centers for Disease Control and Prevention, the suicide rate among people aged between 15 and 44 are somewhat flat while those who are aged 44 or older more likely commit suicide. We ignore those who age from zero to 14 because suicide is a rare case. So we use “percent of population aged between 15 to 44” in each state as one of the independent variables, and we expect that the more people aged in this range, the smaller the total suicide rate should be in a given state (the coefficient should be negative).
Gender disparity (% of male population), calculated as Surprisingly, it is found out that there is a large disparity between male and female regarding this topic. Males take their own lives nearly as four times the rate of females. So we decide to use “percentage of male population” in each state as a regressor, and we expect that the suicide rate is higher in states with higher percentage of male population (the coefficient should be positive).
Income impact (logarithm of real average income), calculated as Apparently, it is much easier for people to feel depressed if s/he only lives around the poverty line, because the basic needs may not be satisfied. Once people are depressed, they tend to have a higher rate of committing suicide. And certainly, states with lower real average income will by statistic have more people living at a low standard. Therefore, we use the logarithm of real average income as a regressor, and we expect that the higher the logarithm of average income is, the lower the suicide rate should be (the coefficient should be negative).
Drinking problem (% of binge drinkers), calculated as Here, binge drinker is defined as people who reported consumption of 5 or more drinks of alcoholic beverages on at least one occasion during the past month. People who are seriously drunk have higher probability of doing something unconsciously or impulsively. They often express their dissatisfaction and rage in some extreme form after binge drinking. So we decide to use percentage of binge drinkers as a regressor and we expect the higher the percentage is, the higher the suicide rate should be (the coefficient should be positive).
Racial disparity (% of minority population), calculated as: In accordance to some study, suicide rate is much higher among minority group such as American Indian/Alaska Natives and Hispanic. So we use “percent of minority population” in each state as one of the independent variables, and we anticipate that the state with higher minority population should be associated with higher suicide rate (the coefficient should be positive).
Economy Impact (unemployment rate (state), national unemployment rate, real GDP per capita) Higher unemployment rate indicates that more people are living at a low level of living standard, which can lead to greater tendency of that people get into self-destruction with depression and gloominess. Therefore, we choose unemployment rate as one of our regressor and we expect that the higher unemployment rate is, the higher suicide rate there should be (the coefficient should be positive). National unemployment rate and real GDP per capita could be the indicator of time change, since we are dealing with panel data. In general, we expect higher national unemployment rate and lower real GDP per capita should lead to higher suicide rate.
Part C: Data
How we got the data:
All data for both dependent variable and independent variables are obtained through the Internet. The sources for these variables vary, and they are listed as reference attached at the end of the paper. All data sources are from government website, organization website or education website. The data all come from preliminary research or census, and therefore the data are credible. Means, standard deviations, maximum and minimum of each variable are reported as below:
Table 1 – Statistic Table for All Variables
Independent Variables Variable Name Mean Std. Dev. Maximum Minimum
Suicide Rate sui_r 12.4146 3.2631 22.0000 6.1000
High school completion rate high 0.8609 0.0385 0.9270 0.7810
% Bachelor’s degree or more bachelor 0.2656 0.0487 0.3760 0.1510
% of people aged between 15 and 44 age 0.4260 0.0148 0.4759 0.3932
% of male population male 0.4923 0.0065 0.5094 0.4812
Real average income ave_inc 31,626 4,908 48,032 22,386
% of binge drinkers drink 0.1503 0.0315 0.2490 0.0810
% of minority population minority 0.1216 0.0977 0.4072 0.0037
Unemployment rate (state) unemploy 0.0511 0.0097 0.0780 0.0330
National unemployment rate natlun 5.4667 0.2949 5.7830 5.0750
Real GDP per capita gdp 38,067 1,568 40,100 36,300 Part D: Regression results Interpretation and Sensitivity analysis
This section presents an analysis of the effect on suicide rate using the data set of 48 U.S. states in 2002, 2004 and 2005.
1) Discussion of the base and alternatives specification. As mentioned in the motivation part, we will focus on the most important factors that could be determinants of committing suicide. According to the research conducted by Centers for Disease Control and Prevention, unemployment rate and education level have the most dominant impact on suicide. This analysis focuses on these primary factors as well as other potential factors that may affect suicide, holding other factors that are beyond our control or unavailable constant.
Base variables：Unemployment rate in each state % High school degree in each state
Alternative variables: National unemployment rate Log (national real GDP per capita) % Bachelor’s degree % age15-44 ％male Log (Average Income) % binge drinker % minority
2) Tabular presentation of result.
Column (1) in Table 2 presents results from OLS regression of the suicide rate on unemployment rate in 48 U.S states without state effect and time effect. As in the cross-sectional regressions in 2002, 2004 and 2005, the coefficient on the unemployment rate is negative and statistically insignificant at the 5% level. According to this estimate, higher unemployment rate triggers lower suicide rate.
Column (2), which includes state fixed effects, suggests that the positive coefficient in regression (1) is the result of omitted variables bias (the coefficient on unemployment rate is 3.426). The Adjusted-R2 jumps from 0.005 to 0.949 when fixed effects are included; evidently, the state fixed effects account for large amount of the variation in the data. However, high Adjusted-R2 is mostly due to the processing of the cross-section data, which introduces 47 more dummy variables.
In column (3), we add national unemployment rate and real GDP per capita in our regression along with state unemployment rate. Since time effect in this case is not explicit, instead of including time as a regressor directly, we use these aggregate factors in economy to implicitly reflect the differences moving together with time, which turns out to be consistent with our expectation. The results in column (1)-(3) are consistent with the omitted factors being important determinants of the variation in suicide rate across states.
Column (4) includes another base variable－% high school graduate. We find out that the coefficient on this variable is negative and statistically significant at the 1% level, which is in compliance with our previous expectation. The results suggests that higher rate of high school completion will lower the suicide rate across states.
Column (5) - (7) include additional potential determinants of suicide rates, along with state effects. As reported in column (5), age disparity (% age15-44) and gender disparity (% male) drive down the effect on unemployment rate across states. However, they are not statistically significant at the 5% level. Even though the estimate coefficients are consistent with our expectation, which is that male and senior people are more likely to commit suicide.
Column (6) includes unemployment rate, % high school graduate, % age 15-44 and % male in 48 states. This regression results the best fit with our expectation, with relatively high Adjusted R2. Recall our motivation, this project aims at finding the most important determinants on individual’s suicide tendency and helps provide numerical evidence for possible decision makers. The regression results suggest that the estimate effects of these individually oriented variables are statistically at 5% level.
Column (7) report the regression that checks the sensitivity of these conclusions to changes in the base specification. The regression drops the variables that control for economic conditions (national unemployment rate and logarithm of national real GDP per capita). In addition, we change % high school degree to % bachelor’s degree and add new alternative variables (% binge drinker and % minority and log(average income) ). However, at 5% significant level, the coefficient of these newly added variables is insigficant. The regressions in column (8) - (10) showed some interesting results. From column (8), we can tell that the negative coefficient on % binge drinker is not a reasonable estimate (as explained before in part B, a positive coefficient was anticipated), which suggests that more binge drinker is associated with lower suicide rate. There must be omitted variables which causes downward bias on % binge drinkers. According to column (9) and (10), % male is one of omitted biased variables. Since the coefficient of % male and covariance between % male and % binge drinker) are both positive, the coefficient of % binge drinker should be upward biased, which means if without this regressor, the coefficient for % binge drinker should be less negative or more positive. However, the regression showed us just the opposite result. Therefore, there might be some other omitted variable(s) which is/are correlated with % of binge drinker.
3) Regression model
To sum up, we find out that regression (6) gives us the most satisfying result:

Interpretation of estimated coefficients:
1) Unemployment rate across states: if unemployment rate increases by 1% point, the suicide rate will increase by 15.887% points per 100,000 people.
2) % high school graduate: If people who get high school diploma increase by 1% points, the suicide rate will decrease by 5.405% points per 100,000 people.
3) % age15-44: If the portion of population aged between 15 to 44 increases by 1% point, the suicide rate will decrease by 16.520% points per 100,000 people.
4) % male: if the portion of male increase by 1%, the suicide rate will increase by 44.702% points per 100,000 people.

Part E: Conclusion
Major findings
Although there are a lot of regression models could be established to explain the state suicide rate in the U.S., the best one we found out says that the state suicide rate in the U.S. is greatly affected by four major factors: state economic environment (state unemployment rate), overall education level (high school completion) and component of population (sex and age group). Analysis in details are provided in part D. In general, the greater the state unemployment rate is, the higher the state suicide rate will be; the higher the high school completion rate is, the lower the state suicide rate will be; the higher the percentage of male in the population, the higher the state suicide rate will be; and the larger the age group 15-44 in the population, the lower the state suicide rate will be.
In conclusion, the result indicates that, if we want to prevent suicide, we may:
1) Pay more attention when: the economy is bad, i.e. unemployment rate is high
2) Improve overall education level, and reduce high school dropout rate
3) Focus more on people who are senior or/and male and provide them with more help and services, such as psychological counseling.
Ways to refine the research
The following things can be done in order to refine the research:
1) Obtain data for other years, and make the whole data set more completed and reliable. We intended to do a regression model using data from a five year period. However, although we tried very hard, we did not find any access to the data of dependent variable in year 2003 or 2001.
2) Find out more potential independent variables. As we discussed in part D, when analyzing the coefficient for % of binge drinkers, we realize that there might be some other important omitted variables which are not included in our current project. We hope that we can find out a reasonable explanation of such situation.

References & Data Sources
Annual Average Data on Unemployment rates for States: http://www.bls.gov/lau/#tables
Educational Attainment: http://www.census.gov/population/www/socdemo/educ-attn.html
Per Capita Personal Income by State: http://bber.unm.edu/econ/us-pci.htm
Percentage of adults who reported binge drinking by state and gender, BRFSS, 1984–2006: http://www.niaaa.nih.gov/Resources/DatabaseResources/QuickFacts/Adults/brfss03.htm
Population by age, sex and race: http://www.census.gov/popest/archives/2000s/
Real per capita GDP: https://www.cia.gov/library/publications/the-world-factbook/geos/us.html
Yang, B., & Lester, D. (2007, April). Recalculating the Economic Cost of Suicide. Death Studies, 31(4), 351-361. Retrieved April 21, 2009,: 10.1080/07481180601187209
U.S.A. Suicide: Official Final Data: http://www.suicidology.org/web/guest/home
Unemployment rate data: http://research.stlouisfed.org/fred2/data/UNRATE.txt