--- In [email protected], "authfriend" <[EMAIL PROTECTED]> wrote: > > --- In [email protected], new.morning <no_reply@> > wrote: > > > > <snip> > > In re-reading the paragraphs, its ambiguous as to whether > > the outlier was actually excluded from the analysis. > > Yeah, I don't think it was excluded. > > > If it was excluded, (which > > was my take upon writing the original post, I do not feel that there > > was sufficient explanation as to why the gang-shooting was so > > extraoridnary out of the ordinary that it should not be included. > > Remember, what the guy is defending against is > the accusation by the critic that the 36-hour spike > in the murder rate meant the whole study was a > complete failure. What he's doing is explaining > why it didn't mean that at all. > > On the other hand, the numbers involved wouldn't be > enough to significantly affect the results even if > the murder spike *had* been included.
But, IMO, each crime type should have been analyzed separated. Average or summing, while ok as a summary device, hides what really happened across the three distinct crime types. And "swallows" murders. > > Second, a key premise of the rebuttal was that temperature "nailed" > > variations in crime. It was said to be a very tight fit to the > sesonal > > crime data. This is key in distinguishing ME from some other factors > > in crime reduction. Yet, with a deeper look, there is still huge > > variations in crime, even after crime has been controlled by > > temperature (and other factors). > > But not as "huge" as the drop in the rate during > the course, right? Well, I am going by the graphs (and commnets in the rebuttal) not the actual data. And they are a bit fuzzy. So to answer your question specifically above, I am not sure. My point is that if there was a doubling one week, and a halfing the next, you have very flucuating data. The more fluctuation, technically, the higher the standard deviation in the data, essentially the larger the change is required to show significance. And the impact, other than in 1-2 weeks, looks pretty modest. And the fact that there was a big drop in cime in 1-2 weeeks and not others is unexplained -- and adds the premise of a highly fluctuating "sea" of crime rates. And a more difficult situation to distinguish an ME. You can see a small object floating on the water in a very calm sea. It disappears when the seas "fluctuate" a lot -- aka are very choppy. > Thus, it points towards other > > possible major factors which have not been controlled for in the > core > > crime model (without ME). This raises serious questions to whether > it > > was ME or other uncontrolled for factors that were driving the > > changes. And/or there simply is a lot of "static", unexplained or > > random variations in the crime rate. In either case, its a difficult > > "base case" from which to clearly isolate an ME during the period of > > its intervention. > > What about the fact that there is normally such a > small number of murders (only 3 percent of violent > crime in 1993 as a whole)? Is it not the case that, > as he says, "with numbers as low as this...random > fluctuations can appear extremely high when listed > as percentages"? His point is valid, but off the main point, IMO. If you average 10 murders a week, and you get 20 in a particualr week, its not an aberation to say murders doubled in that week. Where his point would have relevance is in murders in a small town like FF. If there is one every five years, and you get one this year, the percentage increase for the year is 500% (if you used the average of 1/5th). Or god forbid, two in one year: 1000% That is a distorted view, IMO. > In any case, I'm not sure in what way your second > objection relates to the point he was trying to > make, i.e., that the spike doesn't affect the > study's conclusions. I was commenting on the 3-4 paragraphs you provided, and asked for comment on, not just that one point. If I got "off point", sorry. But I don't see where. > > Third, the rebuttal reinforces the fact that murder, rape and > > assualts, were summed. These are qualitatively different acts. At a > > minimum, separate analysis of each type of crime should have been > > analyzed. If only one or two types of violent crime went dowm and > the > > other(s), did not, it raises quesions as to why, and IMO, would > place > > doubt on ME. By averaging, with assaults being by far the highest > > category, the study becomes essentially a study on assaults. Effects > > on murders and to a lesser degree, rapes, are submerged. > > This makes sense to me as a criticism of the > study design. Why they chose to lump all violent > crime together isn't entirely clear, but I can't > imagine they did this because they *expected* the > murder rate to spike and *wanted* to submerge it. I am not sure pooling was part of the pre-study study design. There can be a lot of leeway if reporting results. I suspect they pooled results because it gave a clearer less ambigous picture. That is, I suspect murder eitherwent up, or stayed "lvel" and possibly the same with rape. That would be harder to expalain than all going down. I don't think they felt bound by any pre-study articulation of thier analysis plan. And it would have been silly to state "we are not going to look at each crime by itself." > I suspect they did it because it wouldn't sound as > impressive, PR-wise, to claim a decline merely in > assaults, as opposed to a decline in all violent > crime; And that murders and/or rapes went up. > and to do studies on all three types of > violent crime separately would just have been too > complicated. No. Its would have been easy. And easier to explain. And reduce a lot of need for rebuttals to criticisms of the study. > My guess is they figured the number of rapes > and murders would be small enough that they wouldn't > significantly affect the results either way in any > case, so they might as well include them. Thats fine to do in a summary, IF they established, seperately, that all three went down. > Anyway, I'm *still* not clear why you think his > explanation for why the spike didn't invalidate > the study, as the critic claimed, lessens his > credibility, I raised three points in the above post, and several more in a prior post, that I thought were weak. Regardless of why they were being raised. > whether the additional murders were > included in the totals or not. I agree/view that, as the analysis is (poorly) structured, the spike in murder rate, and whether is its included or not, does not invalidate the study as done. I also view that because of the poor analysis approach, pooling crimes, etc. the study is flawed, weak, and indicative of "hiding" unpleasant results. > (BTW, thanks for keeping your response simple and > relatively jargon-free!) Sure, I am trying. Its a razors edge sometimes -- of sounding silly to some, and incomprehensible to others.
