Concealed Carry Permit Holders are One Third as Likely to Commit Murder as Police Off

[Kleanbore]

I have already commented on this point in Post #5, but I contend that no one in his right mind would consider the sample size(s) to be statistically valid.

 

[Mike1234567]

Perhaps not statistically "valid" but certainly statistically "interesting" and worthy of further investigation. If further investigation finds a substantial increase per capita of police domestic violence/murder then the next step is to investigate "why". Does it have to do with stresses? Training? Management/politics? Frequent observation of violent acts by others? Perhaps these officers were forced to shoot at perps at some time in their career? Is it a power thing? What?

 

[Frank Ettin]

I have already commented on this point in Post #5, but I contend that no one in his right mind would consider the sample size(s) to be statistically valid.

And we have GEM's comment in post 25:

Where is the significance test and the effect size analyses?

Meaningless without such. Can the OP provide those?

GEM is an academic who writes and has published scientific papers in which he must use statistics in a formal, rigorous way. His methodology must pass the scrutiny of other professionals.

Having worked with experts to both present and challenge opinion evidence based on various statistical analyses, I also have some sense of the importance of being able to test the sensitivity and confidence of such analyses.

For example, especially when dealing with small sample sizes, an analysis is very sensitive to the time period selected. Here the time period 2008 - 2011 was used. Is that really a long enough period? Testing other periods within that range would help demonstrate how sensitive the result is to the time period selected. Testing longer periods, and shorter periods within those longer periods would also test the sensitivity. With small data sets, small variations can make big differences.

And we still have the fundamental flaw that the OP's conclusion isn't even supported by his data selection criteria:

1. The OP's conclusion (thread title) claims that police are more likely to commit murder than CCW holders.
   
   
2. But murder is a particular type of criminal homicide, i. e., the intentional killing with malice of one person by another.
   
   
3. But the OP didn't look at all murders. He looked at only what he calls "domestic homicides." He defines a "domestic homicide" as (post 75):

   ...the illegal killing of an immediate relative, family member, boyfriend or girlfriend, or former boyfriend, girlfriend or ex, and others killed as part of the incident.

   
   
   • "Illegal killing" would include voluntary and involuntary manslaughter.
     
     
   • The OP looked at only particular victims.
   
   
4. So while the OP claims in his conclusion to tell us something about the comparative rates of murder (killing with malice of anyone) he measured, according to his statement of his criteria, all illegal (murder, voluntary manslaughter and involuntary manslaughter) killings of a particular class of people.

Perhaps not statistically "valid" but certainly statistically "interesting" and worthy of further investigation....

The question may well be interesting and worthy of further, and proper, investigation. But that does not justify publishing erroneous conclusions based on questionable data and methodology.

 

[Dean Weingarten]

The type of statistical tests that have been asked for, significance and effect size analysis, do not seem appropriate to the subject.

The article is not a statistical experiment that is attempting to determine cause and effect with a control group. No cause or effect has been postulated.

It is a simple observational study that attempts to determine a ratio of the frequency of events.

An analogy would be to try to determine the ratio of the number of days when it rained between Arizona and Wisconsin, by searching for news accounts of rain in publications for those two areas, that are gathered by weather aficionados who are dedicated to gathering news accounts of weather events.

One year would give an estimate, but there could be a drought in one state and a record year for rain in another. Four years gives a better estimate. I expect the ratio that is considered in the study will be refined as data builds up. The analogy is not complete, in that the news reports are not for separate areas, but for separate populations that are about the same size, one national, the other in one state.

It will vary year to year. However, if, from the very start, the reports show that there are more days of rain in Wisconsin than there are in Arizona, and all of the years tabulated have been shown to have more rainy days in Wisconsin than in Arizona, it is reasonable to believe that it rains more often in Wisconsin than in Arizona, even thought the results are not applicable to statistical analysis for significance and effect size analysis.

In the data link I ask for observers to forward any news items that would add to the data that I already have. Ten years from now we will have more. There may be better sources of data that I have not found.

 

[Dean Weingarten]

Frank Ettin posted:

The OP's conclusion (thread title) claims that police are more likely to commit murder than CCW holders.

That is a mischaracterization. A thread title is not a conclusion. The conclusion was clearly stated in the final paragraph, which you quoted part of:

There are no complete and definitive sources of data that will give us an accurate ratio of unjustified homicides committed by police compared to CCW holders. The numbers are very small and no one keeps a national record of them. However, the numbers found for domestic homicide cases, which are some of the easiest solved and most highly publicised cases, offer strong evidence that CCW permit holders are less likely to commit unjustified homicide than police officers, as little as one third as much.

 

[Frank Ettin]

...The article is not a statistical experiment that is attempting to determine cause and effect with a control group. No cause or effect has been postulated.

It is a simple observational study that attempts to determine a ratio of the frequency of events...

Garbage. A properly conducted observational study is a statistical experiment, and it's validity is dependent on statistical principles.

Indeed, you have entitled your blog entry:

Concealed Carry Permit Holders are One Third as Likely to Commit Murder as Police Officers

And based on your "observational study" you can not truthfully say that. You determined no such thing.

Your blog should be entitled:

1. Based on the data and methodology discussed below, during the period 2008 -- 2011 Florida CCW holders committed unlawful killing (i. e., murder (criminally culpable killing with malice) or a form of manslaughter (criminally culpable killing without malice)) of a family member or intimate partner (i. e., relative, spouse or former spouse, boyfriend or girlfriend, or former boyfriend or girlfriend) at a rate approximately one-third that of sworn, full-time police officers nationally.
   
   
2. However, because --
   
   
   • Only the specific unlawful conduct was examined; and
     
     
   • The data set is small; and
     
     
   • No data outside the subject period was examined; and
     
     
   • No effort was made to test the sensitivity of the data and analysis to the study time period selected;
   
   no conclusions may properly be drawn beyond the specific conclusion described above.

Frank Ettin posted:

The OP's conclusion (thread title) claims that police are more likely to commit murder than CCW holders.

That is a mischaracterization. A thread title is not a conclusion...

The thread title appears to be a conclusion, will be understood by readers to be what your study proves and concludes and even as a mere title misrepresents what your study shows.

The truth is that you have absolutely no grounds upon which to claim that police officers are three times more likely to commit murder than CCW holders. Nonetheless that is what you claim in the title of your blog.

 

[Dean Weingarten]

Then we will simply disagree.

Inference from a subset of data is a common and well understood practice. Data is commonly incomplete. If only 14 rain events have been reported in one place for four years, and 52 in another, it is a good indication that one place has three times as many rain events, even if the only events that are reported are those of 1/4 inch or more. It is not proof, I will grant you that, but the nature of some rare phenomena are by themselves not subject to the type of statistical analysis that you demand.

 

[Kleanbore]

Posted by Dean Weingarten: Inference from a subset of data is a common and well understood practice.

One of the things that are well understood about that practice is the need for adequate data.

If only 14 rain events have been reported in one place for four years, and 52 in another, it is a good indication that one place has three times as many rain events, even if the only events that are reported are those of 1/4 inch or more.

How good is the question. I would not make an investment decision or base any other conclusion on it.

But that's not a good analogy. Rain or no rain involves a binary outcome. The homicide data cited involve how many persons died in aggregate, and that brings in such factors as family size, how many persons happened to be present at the time of each incident, how many were able to escape, and so on, variations among which could very easily alter the outcomes.

 

[GEM]

It is a simple observational study that attempts to determine a ratio of the frequency of events.

Why are there numerous statistical techniques that have been developed to analyze frequency of events?

Chi-square, logistic regression, discriminant analysis, etc.

Why did I get sent to an SPSS seminar dealing with the proper way to deal with large data sets from observational methodologies? Design, use in prediction were covered.

If you want to say this ratio predicts something, then you need such.

 

[Frank Ettin]

Then we will simply disagree....

So you continue to clam that you proved that, "Concealed Carry Permit Holders are One Third as Likely to Commit Murder as Police Officers"?

...Inference from a subset of data is a common and well understood practice. Data is commonly incomplete...

Yes it is. And that is why, as GEM points out, the common practice is to test such inferences using various statistical tools. Without such testing claiming the validity of such inferences is intellectually dishonest and methodologically unsound.

 

[jrdolall]

Why is this thread still open?

 

[Robert]

Good question. Seeing as there is not much of any use coming from leaving this open I am going to close it down.