How exactly is this a problem?
Hi All,
I am running a fixed effects model in stata, but all my independent variables are statistically significant. I am wondering if I am doing something wrong because I have never seen this happen before. Has anyone ever encountered this problem before?
Last edited by royalstatus; 09-23-2013 at 05:30 PM.
How exactly is this a problem?
I don't have emotions and sometimes that makes me very sad.
I'm not particularly sure if it is a problem. I have about 19 independent variables, and they are all statistically signficant
Are any of these paradoxical results. How did all of the variables fair in bivariate analyses? You need to provide much more information in order for us to explore this topic with you.
Stop cowardice, ban guns!
I'm going to peer into my crystal ball and say.... you have a large sample size. What say you?
Why don't they never ever mention their sample sizes
when describing their problems? That is very strange.
Probably because starting statistical text and classes never mention power.....or sample size.
None I read or attended did.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
That is so blatantly and obviously wrong.
Every introductory class I ever took or served as TA for (across 3 different universities, in psychology departments and statistics departments) talked at least a little about statistical power.
And please point to one introductory statistics text that never mentions statistical power. Really-- even one.
In God we trust. All others must bring data.
~W. Edwards Deming
I suspect you took statistic classes in a statistics department jake. Statistical classes, at the undergraduate level, in other programs such as social science are very different.
As for text look at the Sage monograph "Applied Regression: An Introduction" by Lewis Beck or Understanding "Regression Assumptions" by William Berry (my graduate introduction to regression) and tell me where either talk about power at all.
For that matter look at John Fox's "Applied Regression Analysis" (which one of my graduate professors made optional because it went far beyond the course) and tell me where it talks about power
That's 3 and all are really beyond most introductory courses particularly Fox.
Last edited by noetsi; 09-24-2013 at 04:27 PM.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
noetsi I would also be interested to see the texts or the university papers you're thinking of. To me, if a teacher is calling their paper a "statistics" paper and not even mentioning issues as basic as this then surely there's a bit of false advertising going on?
noetsi, as I mentioned in my previous post, many (most actually) of the classes I mentioned were in psychology departments and not statistics departments. All were at the undergraduate level.
As for your your text. Fox clearly does talk about statistical power in various different sections, as a search of the text on Google Books plainly reveals. As for the other "text"... it is 79 pages long! What do you really expect in a pamphlet of that size.
In God we trust. All others must bring data.
~W. Edwards Deming
Those were the regression books I used in my graduate classes for stats.
As for Fox you are welcome to show me any line in that book (which again is well beyond what most students see in their stats classes in social science) where he suggests that greater power will increase the chances of rejecting the null. I went through the indexes of the book and reviewed some of the chapters and saw nothing like it. Power is not even in the index or any of the chapter subtitles
Except maybe for economics psychology has far more statistics than is common in social science programs. Most students are not going to be exposed to that level of statistical discussion even in grad programs. I was not and I went through three before my statistics program (where we did talk about this issue a lot).
I googled the book and statistical power (and just power) and found no references. If you could send me the links you found this discussed in I can admit I am wrong
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
Beyond what victor and I already said, power when presented gets discussed primarily in terms of a type I and II error which many who don't have a good statistical background don't connect at all to the fact that there parameters are signficant when they probably should not be. I understand they should understand how these relate, but I know from my own experience it goes right over their head. Because it went over mine at the time.
Much of stats courses early on focus on theory and not such practical elements. Or at least they make it more complex than students brand new to stats can make sense of.
"Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995
At my university I did a 100-level stats for business course early on, and while I don't think it explicitly used the word "power" they certainly talked about how the determinants of a p value include sample size. The general research methods for psychology undergrad papers also talked about the importance of sample size at least conceptually. Then power was explicitly covered in at least one of the graduate research-methods-in-psychology classes I took (I remember one of Jack Cohen's articles being part of the course material). And my dept is not one that pushes quant methods very hard (we have a lot of qual researchers). So it's confusing to hear that this wasn't covered in your courses
we didn't even have any topics on type I and type II errors... or sample size calculations... as a matter of fact the statistician who taught those courses did not know them himself! He never used any sample size formula in any of his theses!! He used to estimate the necessary sample size from his gut (seriously) and people used to believe in him like god!! Later when new statisticians were hired by university, our students saw "sample size calculation" (at basic levels). That old statistician still estimates sample size by his gut though! By the way he was chosen as the national researcher of the year a couple of years ago. True story!
the statistician who ruined my thesis was someone else. He knows much more than the above person and still doesn't know the importance of power.
"victor is the reviewer from hell" -Jake
"victor is a machine! a publication machine!" -Vinux
Tweet |