In both domestic and international
affairs, crime rates often serve as an indicator of the overall level of safety
within a nation, and lowering crime rates is an important aspect of protecting
the populace. While many factors may contribute to the level of crime, this
analysis examines the relationship between crime rates and regime type.
Isolating the effect of regime type on crime rates contributes to an
understanding of whether crime is more prevalent in democratic or autocratic
states, indicating which countries might need more resources to enhance law
enforcement effectiveness or to pursue other measures to keep the populace
safe.
Little analysis of crime rates at a global level exists. Gene Stephens suggested that democracies would experience higher crime rates, as these nations allow dreams of prosperity with insufficient avenues towards that goal. However, Stephens also highlights the “democratization of governments,” or governments in transition, as a cause of higher crime rates, leaving room for debate about the differing effects of established and changing regimes on the level of crime.[i] In contrast to previous research that examines changing entities, my analysis includes existing regimes.
I employ multiple regression analysis
to examine the effect of regime type on crime rate.[ii]
I also include three control variables that might affect the
relationship between regime type and crime rates. Socioeconomic factors,
education level, and the presence of conflict could correlate with both regime
type and crime rates. I use GDP per capita as an indicator of socioeconomic
status. Wealthier countries might have less crime and be more likely to be
democratic, and higher levels of education may have a similar effect. Finally,
the existence of conflict within a country may result in more crime and can
affect the democratic or autocratic nature of a regime. Compiling these data yielded a dataset of 95
countries with scores on each of the independent variables.[iii]
I began by
testing the variables for skewness (whether
the variables were symmetrical or had extremely high or low values that could
affect the analysis) and multicollinearity (when overlap between
variables changes or negates their effects on the dependent variable).[iv]
I then regressed crime rates on the independent variables. Figure A is a visual
depiction of the effect of regime on crime rates. As shown in the graph,
as regimes become more democratic, crime rates increase. However, I find high variance among countries with extremely high democracy scores. It is possible that the high crime rates are found in states that are newly democratized or completing transitions, resulting in high levels of crime. The graph suggests, however, that countries at the highest levels of democracy may have lower crime rates, as shown by the multitude of level-10 democracies with almost no crime. It should also be noted that the low crime rates of highly autocratic
regimes may be related to inaccurate reporting by those regime types.
Table 1
presents the estimates of the variables. Model 1 of the table depicts the
relationship between regime type and crime rates without the control variables.[v]
In this model, a one-unit increase in regime is associated with a 2.6%
increase in crime rates, on average. In the multivariate regression, shown in
Model 4 of Table 1, the effect of regime type is higher; for a one-unit increase in regime
type, crime rates increase by 8.8%, on average and controlling for other
variables. The effect in Model 1 is not statistically significant; however, when other variables are included in the model (as shown in Models 2-4), the effect of regime type on crime rates is statistically significant, with a p-value
of less than 0.001. This value indicates that there is less than a 0.1% chance
that the observed relationship is due to chance. Additionally, in Model 4, which includes all the variables, both the education and GDP variables are statistically significant. On average and controlling for other variables, a one-unit increase in the education index is associated with a 9.7% decrease in crime rates, and a one-unit increase in GDP per capita is associated with a 4.1% decrease in crime rates. This model
explains about 31% of the observed variation in crime rates.
Table 1. Determinants of Crime Rates
An
examination of Romania’s regime type and crime rates between 1986 and 2012 further
affirms these results. Between World War II and the overthrow of dictator
Nicolae Ceausescu in 1989, Romania was highly autocratic, scoring -8 on the
polity scale from 1986 to 1989. Romania then experienced a period of
transition, stabilizing as a democratic state in the late 1990s. It
subsequently joined NATO in 2004 and the European Union in 2007.[vi]
Figure B depicts this transition. Additionally, Romania’s crime rates for this
period are shown in Figure C. The number of intentional homicides rose sharply
during the transitional period then decreased substantially as Romania reached a high level of democracy.
Thus, both
the regression analysis and an examination of Romania’s data indicate that
while countries transitioning from autocracy to democracy generally experience higher crime rates,
states at the highest levels of democracy may actually have lower levels of crime. However,
the analysis does not confirm causality. Other variables that pose measurement
difficulties, such as law enforcement institutional strength, may also affect
crime rates. Additionally, the data were limited geographically; little data on
crime rates for Africa, for instance, existed. Despite these limitations, the
analysis supports Stephens’ suggestion that democratization would correspond
with higher crime rates; however, the data do not fully support Stephens’ assertion that
democracies would also experience high levels of crime. While some high crime rates in highly democratic countries exist, it is possible that these result from recently transitioned countries. Instead, it may be that states at the highest levels of democracy actually experience lower crime rates.
Appendix A. Summary Statistics
Appendix B. Additional Information on Variables
Appendix B. Additional Information on Variables
[i]
Stephens, G. (1994). The global crime wave. The
Futurist, July-August 1994, 22-28.
[ii]
The outcome variable, crime rate, is represented by the indicator Intentional
Homicide Rates.
[iii]
The data are from the year 2007, as that year yielded the most complete
dataset.
[iv]
I found positive skew in the GDP variable, which is thus logged to correct
skewness. Education and GDP per capita shared some collinearity, but not enough to merit exclusion from the model.
[v]
The squared regime type variable is included to demonstrate the parabolic
effect of the data.
[vi]
The World Factbook 2013-2014.
Washington, DC: Central Intelligence Agency, 2013.
Definitely an interesting topic and one that I haven’t read about much before, so overall I think you did I good job. There are a few things you could change though.
ReplyDelete- Jargon
o While it is sometimes hard to avoid, and you do define some of them, skewness, multicollinearity, confounding, can probably be broken down into simpler terms. The definition (whether the values of the variable were distortionary) is not very clear to those that are not in the subject. Simplifying the language for skewness and multicollinearity may also help in showing the reader as to why its relevant to the argument/study. I know what you’re correcting for or attempting, I’m just not sure how much a general reader would understand.
- Analysis
o I think most of it is good, but when you refer to the graph as showing a relationship between high crime and democratic regimes, the lowess line appears to curve downward around the values of nine and ten. So while the line does possess a positive relationship, this part seems to need some explanation. Maybe that part is connected to GDP.
o The explanation of the impact of GDP having no effect is somewhat confusing since it shows up as statistically significant in models 2-4 unless I’m reading something incorrectly. Clarification might help out.
- Table Presentation
o I would recommend adding a few borders to your tables just for presentation sake, much like the ones we have to duplicate in the problem set. That might just be personal preference, but I appreciate the border between the DV section and the IV… and the Constant and R-Squared.
Hope this helps
I think this is a really neat topic, especially given all the intricacies that come with measuring regime type and subsequent criminal activity. I think one issue that you may be running into is that certain regime types are more likely to skew or misrepresent their crime rates and generally lack transparency. This may make your results bias and actually just be presenting better reporting by democratic states. I'm a little confused about Stephens' argument, why is it exactly that he thinks democracies have higher crime rates? Are you saying that there's more freedom in society so people are less deterred form committing crimes? I'm also curious as to why you're using education as a control variable. It seems to me that the relationship between education and regime type might be spurious, although I definitely see the correlation with crime rates. In regards to presentation, I would maybe add some "pretty" (Hart, 2015) to your graph with colors etc. In terms of your analysis of GDP not having an effect, you're probably thinking that because the coefficients are 0.000. However, the variable measures big numbers and it actually has statistical significance, so even the smallest coefficient would have substantial implications. Maybe you just need to extend the decimal out, especially since it has a negative sign which points to some sort of a relationship. Overall, good job!
ReplyDeleteHey Jenny,
ReplyDeleteI think you did a great job in terms of structure and language, the post is very clear and concise. I can see Mike’s argument about some of the terms being a little technical, but would say that you explained them very well (except maybe where you use “distortionary”, but you probably shouldn’t count on my English skills at that point).
Like Natalie, my only concern regarding clarity is with the passage where you introduce Steven’s argument: Is it about regimes “in transition” or “consolidated” regimes? If it is the former, it would indeed be a completely different question theoretically, and I would recommend that you emphasize this a bit more (“previous research has looked at … but I would like to focus attention to....”).
The selection of your control variables is intuitive and to my limited knowledge theoretically well founded in the literature. I would suggest though that you either include or mention at the end “state weakness” in terms of security apparatus, rule of law and administration/taxation. Most state fragility indexes may include crime rate (I haven’t checked), which would make it problematic to use them without transformations due to collinearity. Since you already mention law enforcement institutions at the end, you might want to just add administrative and policing capacity to that list.
I suggest this particularly because I think that the presence of an actual (intrastate?) conflict may not be necessary to increase crime rates: A states control over its territory might be weak in any of the above terms without an actual intrastate conflict being present, (theoretically) increasing the probability of violent crimes (i.e. homicide) on its own. Conflict in that context is a proxy variable (something you could add in a footnote).
Your analysis is in almost every respect both very accurate and carefully interprets the results. In addition to what Mike brought up about the graph, I would personally show less certainty that the relationship is positive (maybe add an “appear to be” or “graph indicates” to the sentence). As you point out yourself at the end, many cases are missing from your dataset. It may be that these are the cases with the highest crime rates, which (frustratingly ;-)) could completely turn around the sign of the relationship you are measuring. On that note, if you have the time, you could give us a graph (distribution) or footnote in the appendix that indicates the regime type of the missing cases. This would also increase the accuracy of your sentence about crime rate in “Africa” being underreported.
This may also be related to the style of Figure A: The maximum value of crime rate appears to be around 5 in your dataset. Maybe the positive relationship shown in your regression table is more clearly visible in the graph if you limit the y-axis ticks from 0 to 5. This could also give you more space to label some of the more extreme cases at the low/top end of the line. In addition, I personally would not mind if you added “(per 100,000 people)” to the y-axis title, even though you mention this in your summary statistics.
Overall, I liked your post very much, since all the important elements where included, you interpreted your results carefully and correctly (except for that GDP-issue), and I felt that your style is impeccable. If you make some minor adjustments as per the comments, I can’t see why Professor Hart wouldn’t agree that this is an excellent blogpost.