Explaining Regression

In trying to simplify something, you’ve made it more confusing. I think the most important thing is to drop the statistics speak altogether–no need to mention variables, correlations and “controlling.”
I’ve explained regression analysis to my math-phobic wife a couple of times and she now seems to understand. The way I do it is by saying something like this (using your weight example): It allows us to create two identical individuals who are alike in every way except for one characteristic, in this case weight. We then compare life expectancies and arrive at a reasonable estimation of the impact of individual weight. To be sure, I’ve glossed over some important facets, but the basic concept is still accessible to lay-people.

Bob: You should be more concerned about analyzing the A’s wins/losses than some economist’s take on their organizational structure. Just saying.

I took it as a measure of Stewart’s interest that Levitt was on for two segments. Most guests are only on for one; it’s usually only the big names that merit two.

But Levitt isn’t trying to prove that every single abortion reduced crime. He’s only trying (successfully, I think) to prove that in the aggregate, legalizing abortion had the effect of reducing crime. It could be that some significant number of abortions were of fetuses that would have grown up to murder people, but some small number of abortions were of fetuses that would have grown up to stop crimes. Nevertheless in that scenario, the abortions in toto would have ended up reducing crime.

So regression is the right statistical tool for the job.

Since I’ve been following medical research reports for decades what I’ve learned is that it is exceedingly difficult to control for variables in large population studies.

Look at the recent report claiming that mild to moderate obesity does not decrease life expectancy. Lots of studies have been done by generations of sharp medical researchers producing contrary results. Did it take decades to finally come up with a way to control for all the variables that could distort the results? Or will we find out a few years hence that the latest study is flawed?

Some economist confidently proclaims that he has picked the signal of abortion’s effects on crime against the background of countless other changes in society. Some other economists think that since this economist is so bright he must be right. Well, sorry, I’ve seen too many very bright medical researchers get things wrong after repeated attempts and I’ve seen the conventional wisdom swing back and forth too many times on other questions that had many more people working on them than this question of abortion and crime. It is at very least premature to proclaim that Levitt got it right.

I happen to think that Levitt is wrong for the reasons Steve Sailer has outlined. But even if Sailer hadn’t made his case I’d still be thinking:
But look at how complex the causes are for crime and how many other debates about crime and guns and crime and broken families have raged for decades without definitive answers. Social science is not that easy.

From what I’d call a scientific philosophical standpoint my problem with Levitt’s conclusion is that he doesn’t offer evidence that proves his vague mechanism. Attempts to control for variables have to do that because this isn’t real experimental science. This is “lets statistically control for variables without the ability to repeat experiments” science.

His mechanism: “Wantedness”? What social science evidence is there that “wantedness” has increased among the children being born? Also, what social science evidence is there that “wantedness” reduces the odds that a kid will grow up not to be a criminal?

An obvious detail about “wantedness” needs to be answered: “wantedness” by mothers or by fathers? Or both? Illegitimacy has been rising since abortion was legalized. So hasn’t paternal “wantedness” been declining substantially post Roe v Wade? His proposed mechanism seems highly dubious.

It is extremely difficult to control for variables when you don’t have evidence for mechanisms on how those variables might influence some target variable under study. Taking Alex’s example of height, under different circumstances with different populations height might be higher because of genetics, better nutrition, or even worse nutrition (high glycemic index diet causing higher average blood sugar causing higher insulin and higher IGF which might cause excess growth) or even use of growth hormones. A genetically heterogeneous population for height might have other genotypes on the same chromosome that also influence longevity independently even though they are linked to the height gene. Or greater height might be correlated with greater symmetry which might in turn be correlated to less toxin exposure and less stress during fetal development and hence tall people might be born with fewer internal flaws on average (though perhaps only in some environments). It is naive to attempt to control for variables without a lot of knowledge about how those variables influence each other and the phenomenon (in this case crime) under study.

liqingchao 07年7月11日

google排名
google排名
wow gold
wow gold
powerleveling
powerleveling
wow gold
wow gold
wow powerleveling
wow powerleveling
wow power leveling
wow power leveling
powerleveling
powerleveling
wow power level
wow power level
world of warcraft powerleveling
world of warcraft powerleveling
world of warcraft power leveling
world of warcraft power leveling
hongkong hotel
hongkong hotel
beijing tour
beijing tour
翻译公司
翻译公司
上海翻译公司

上海翻译公司
北京翻译公司
北京翻译公司
北京搬家公司
北京搬家公司
货架
红外测温仪
红外测温仪
超声波测厚仪
超声波测厚仪
超声波探伤仪
超声波探伤仪
频闪仪
频闪仪
涂层测厚仪
涂层测厚仪
电火花检测仪
电火花检测仪
google排名
仓库
仓库
仓库出租
仓库出租
物流园区
物流园区
集团电话
集团电话
四环素牙
口腔常识
口腔常识

口腔医生
口腔医生
网站设计
网站设计
多媒体
监控
监控
监控系统
监控系统
门禁
门禁
门禁系统
门禁系统
搬家公司
搬家公司
条码打印机
条码打印机
牙周炎
牙周炎

摆闸
痤疮
痤疮
色斑
色斑
黄褐斑
黄褐斑
青春痘
青春痘
口腔医生
口腔医生
口腔
口腔
口腔医院
口腔医院
口腔科
口腔科
三星集团电话机
呼叫中心
电话交换机
交换机
呼叫中心
测厚仪
测温仪
停车场
道闸

cheap Silkroad gold
cheap Silkroad gold
Silkroad gold
Silkroad gold
Silkroad Online gold
Silkroad Online gold
sro gold
sro gold
Archlord gold
Archlord gold
cheap Archlord gold
cheap Archlord gold
Archlord Online gold
Archlord Online gold
buy cheap Archlord gold
buy cheap Archlord gold
Dofus kamas
Dofus kamas
cheap Dofus kamas
cheap Dofus kamas
Dofus gold
Dofus gold
Dofus money
Dofus money
Rappelz Rupees
Rappelz Rupees
Rappelz gold
Rappelz gold
cheap Rappelz Rupees
cheap Rappelz Rupees

wulvyou 07年7月18日
google排名
google排名
wow gold
wow gold
powerleveling
powerleveling
wow powerleveling
wow powerleveling
wow power leveling
wow power leveling
world of warcraft powerleveling
world of warcraft powerleveling
wow power leveling
wow power leveling
cheap wow powerleveling
cheap wow powerleveling
hongkong hotel
hongkong hotel
beijing tour
beijing tour
翻译公司
翻译公司
上海翻译公司
上海翻译公司
北京翻译公司
北京翻译公司
门禁

集团电话
集团电话
货架
货架
货架
google排名
条码打印机
机票
搬家公司
北京搬家公司
仓库
仓库
仓库出租
仓库出租
物流园区
物流园区
条码打印机
猎头
猎头
虚拟主机
红外测温仪
红外测温仪
超声波测厚仪
超声波测厚仪
超声波探伤仪
超声波探伤仪

频闪仪
频闪仪
涂层测厚仪
涂层测厚仪
电火花检测仪
电火花检测仪
beijing
beijing
china tour
china tour
great wall
beijing travel
beijing travel
搬家
搬家公司
搬家公司
北京搬家公司
北京搬家公司
beijing
beijing
china tour
china tour
great wall
beijing travel
beijing travel

摆闸
痤疮
痤疮
色斑
色斑
黄褐斑
黄褐斑
青春痘
青春痘
三星集团电话机
呼叫中心
电话交换机
交换机
呼叫中心
停车场
道闸

China Tours, Travel China, Beijing Tours ，China Travel is a China travel agency providing China tours、Beijing Tours, Tibet Tours, Great Wall Tour and Yangtze river cruises. We specialize in “Tailor-Made” and “Customized” private and group tours to China. We’ll try our best to offer you China tours with super value and most considerate service.

水虫スカルプ育毛シャンプー育毛岩尾シャンプー宮迫シャンプー発毛シャンプー抜け毛シャンプー薬用シャンプー包茎包茎治療包茎手術冷え性むくみ蟹毛蟹ズワイ蟹タラバ蟹産後わきが

住吉出会い平野出会い堺出会い貝塚出会い冨田林出会い箕面出会い東大阪出会い出会い系出会い無料出会い系の攻略について出会いカフェ出会いパーティー優良出会い系サイト