Hello

I think that that you should add gender, violence cases per year, unemployment 

these variables are more important than others.

I think the GINI coefficient and the nemployment rate may interact since the  same level of unemployment may have different effects in more and less equal societies.  I under

I accidentally pjblished the above before I had completed my message.  To continue, I understand the difficulty of what you are trying to do, but wonder about  its feasibility and how well motivated your model is theoretcally.  For example, why is "trade" in the model.  Is it designed to proxy for something?  If so, what?  Also why should government expenditures across 6 regions help hen there can be so much variation within regions, especially on a per capita basis?You don't say what are the units for your dependent variable, but i I am assuming that it is the rate ffor each country, but if so how can you be sure of consistent definitions and reporting by each country?  Also some variables that might make sense are not in the model, such as police per cvapita, army size, governmental type, tribalism, recent civil war, etc.  I don't know how inclusive your list of countries is, but it

....thanks Richard and Arnon for your concise contributions. I'll shed more light on my research. I'm working on the determinants of income inequality and the relationship to homicide rate. To support Richard's argument, I have since removed the 'Trade' variable. Govt Exp is there to know the impact of govt spending in the crime regression. Perhaps, increased spending could increase household income, reduce the inequality and thereby reduce the homicide rate. I also failed to mentioned that I have the 'death penalty' as one of my variables. I'm using the FE estimation technique which takes care of the 'unobserved country-specific' fixed effects. Interacting the Gini index with unemployment makes so much intuitive sense with your explanation....never thought of it......thank you!.....I won't mind more suggestions, though!

I'm not sure what happened but again my writing was truncated.  One portion that is missing is that I suggested that you might want to remove some countries from your analysis if you determine that information on too many variables is missing or unreliable.  This determination should, however, be made before you proceed with your analysis on a priori empirical and theoretical grounds.  Also once you have built your model I suggested that you take a series of random subsets of countries in the model and see how well your global solution fits each subset.  If the overall model is valid, the fit should ordinarily be close with errors randomly distributed, and such testing will add significantly to the confidence one would have in your results.  If different models better characterize a much larger than expected portion of your random subsets, you may gain a clue that will help understand what is going on and allow you to improve your overall model.  On your death penalty variable, there are numbers of ways to specify it, including its legal existence, its application (which can be as a rate or as a raw number of executions) and the kinds of cases to which it can be applied (e.g. ordinary crimes, homicides, treason.)  If, for example, a country has the death penalty but it is only available to punish treason, there is little reason to think it will affect the crimes you are examining.  If you are developing separate equations for different crimes then the death penalty might be specific to those crimes.  There is also a question of whether the DP variable belongs in the model at all given how few developed Western countries have it and how unusual those with the DP (U.S., Japan) are as well as its presence in Muslim countries and the colinearities that will exist.  In terms of variables to add, a woman's education variable might make sense since other research suggests the level at which women are educated relates to numbers of social phenomena; birth rates for example.  A different proxy for female status might be the poportion of government employees who are female.

I am very interested in your data set - what variables are you able to quanity and how are you wanting to explore their relationship/associations - generally it would be good to a theory of crime that can underpin the level of inquiry you wish to make - having worked in the area - ou theoretical permise greatly influences our articulation

Hi J-F,

Homicide rate is the dependent variable and there are 9 explanatory vars: Gini index, GDP per capita, male youth 15-24yrs, pry enrol rate, sec enrol rate, rule of law (rol), corruption, male unempl and death penalty. Out of these, only death penalty is 'binary'. 1 for countries upholding the dtpen and 0 otherwise. I have a global dataset - 137 countries, 7 regions. Analysis from data show that LAC region has higher homicide rate and higher income inequality while Europe is low in both variables relative to Sub-Saharan Africa (base region). I'm just wondering how the relationships among these vars can be further investigated beyond the a priori that the 'Gini index is a strong determinant of homicide rate'. Will it make any intuitive sense to explore the role of institutional quality by interacting rol and unemployment rate; rol and education vars?....any thoughts, please? Thank you.

My understanding is that you want to create an interaction term (s). However, your question is actually two-folded: an interaction term and a collinearity issue. It appears that you want to introduce Rule of Law Index and Corruption Index, possibly both from World Bank’s World Governance Index (WGI). Theoretically, Rule of Law and Corruption Index are highly correlated because a country with a good rule of law is highly likely to have a low corruption.  You can check it by conducting a bivariate correlation analysis between those two variables.

An elevated correlation may also exist between GDP per capita with either rule of law or corruption index. A highly developed country in general has a high level of rule of law, while having a low level corruption.

Another concern may be collinearity between those variables. You can conduct preliminary OLS regression analysis and check variation inflation factors for those variables as well as the bivariate correlation among them. If you confirm a collinearity issue, you cannot introduce those variables in a same regression model. Instead, you can introduce them in separate models.  Alternatively, you can create a combined index out of those two variables: rule of law and corruption index, if it is theoretically supported.

If you are indeed using World Babk’s WGI, you need to be aware that one element of Rule of Law is government’s ability to enforce the law. It is possible that a citizen perceives a high level of rule of law in his or her country when the country experiences a low homicide rate.  In other words, there may be a causal direction problem between rule of law and homicide rate.

Another variable requires an attention is death penalty. Some countries have death penalty laws, but they have not executed for the last few years and abolished death penalty by practice. How did you classify death penalty countries and non-death penalty countries? For example, should the country, that has not executed death row inmates for the last three years, be considered non-death penalty country?  

Additionally, there are many existing studies that already introduced the variables you mentioned. For example, an abundance of cross-national studies tested the relationship income inequality and homicide rate. Thus, I strongly suggest that your research must be driven by theories. Please thoroughly read previous research on the subject and make an adjustment in your research design, including the independent variables.       

The major reason for the act of Homicide is buried in the primary units of environment; it could be affected by home- environment in younger ages, or by the closest community, in teenages. Regression could be trigered by an small spot of "hope" or progress in a case. consequently, ageing would be beneficial for the maintaining in this avenue. Another important element in regression of homicide would be being accepted by bigger scale of a community, an environment or a culture than before, to motivate the sense of belonging to a positive- majority.