www.philosophiaupjp2.fora.pl

cheapbag214s

How much variation is necessary for regression analysis,[link widoczny dla zalogowanych]
Let's say that I'm trying to find the factors which help predict whether I wear red shoes or not during a year long time-period. I could look at weather,[link widoczny dla zalogowanych], day of the week,[link widoczny dla zalogowanych], month,[link widoczny dla zalogowanych], and other factors. However,[link widoczny dla zalogowanych], how much variation does there need to be in the dependent variable for me to be certain that I am acctually measuring something that is meaningful. Do I need to have worn red shoes 2 times? 5? 50? 100? etc?
Basically,[link widoczny dla zalogowanych], how much variation in the dependent variable is necessary to develop a statistical model that generates somewhat meaningful results?
You seem to be equating level of variation with sample size,[link widoczny dla zalogowanych], and that is worth looking at more closely. "I'll call my result meaningful if it explains at least 3.2% of the variance in my outcome in a way that is significant at the .001 level." Then it becomes a straightforward matter of calculation to determine the sample size necessary to achieve a certain level of power (ability to demonstrate a statistically significant effect if one exists). The open-source program GPower is effective for using this type of approach across many different statistical procedures.
+1 rolando2 is right that what constitutes a result is a theoretical question for the researcher,[link widoczny dla zalogowanych], not a statistical question for the analyst. In light of his comments about power,[link widoczny dla zalogowanych], note that the further you get away from 50%,[link widoczny dla zalogowanych], the more data are necessary to clearly resolve the relationship--rare phenomena are simply hard to study. Furthermore,[link widoczny dla zalogowanych], the problems of low probabilities are compounded by the existence of many predictor variables. (Although,[link widoczny dla zalogowanych], 150,[link widoczny dla zalogowanych],000 observations is likely to be more than enough.) gung Jan 10 '12 at 4:07
相关的主题文章：

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]

[link widoczny dla zalogowanych]