Can small differences in a variable (which is being controlled for) not be picked up in a multiple regression analysis?

#1
Hi everyone,
I am reading some research looking at the association between calories consumed at different times of the day and likelihood of obesity. The authors state that they have adjusted the results for total calorie intake over the day. However, they then go on to say that a possible reason for the increased likelihood of obesity amongst people who eat later at night is that they consume more calories over the day.
I'm wondering how this hypothesis makes sense if they adjusted for total daily calorie consumption in their study? Is it possible that the multiple regression adjusts for total calorie consumption but doesn't pick up small differences in calories day-to-day? I.e. the difference in daily calories is too small to be noticed by the multiple regression, but large enough to contribute to an increased risk of obesity over a 3.5 year period?
Thank you!
 

noetsi

Fortran must die
#2
I am not sure how those statements conflict. If you tested, controlled for, overall calories and you say that you discovered that overall calories matter where is the conflict?

A regression model would not pick up or not pick up small differences in caloric intake. The instrument used to gather it might not capture this, but that is a different issue than the regression catching it. Because of issues such as statistical power small effect sizes might not be captured by the statistical model itself (actually I think this would effect the p score not the actual effect size).
 
#3
I am not sure how those statements conflict. If you tested, controlled for, overall calories and you say that you discovered that overall calories matter where is the conflict?

A regression model would not pick up or not pick up small differences in caloric intake. The instrument used to gather it might not capture this, but that is a different issue than the regression catching it. Because of issues such as statistical power small effect sizes might not be captured by the statistical model itself (actually I think this would effect the p score not the actual effect size).
Sorry, I think I used the word 'controlled' incorrectly. It was a cross sectional study so participants' calorie intakes were not controlled. Variations in their calorie intake were only adjusted for in the analysis afterward. I should clarify my question-

If the statistical analysis tells me that there is eating later at night significantly increases the risk of obesity, even after adjustment for total calories consumed throughout the day, then doesn't that mean that you can rule out 'total calories consumed throughout the day' as a variable affecting people's increased risk of obesity? I.e. there must be another reason why people who eat later at night are more obese?

I'm no stats expert so I might have this completely wrong! Thanks for your help.
 

noetsi

Fortran must die
#4
No you can't rule it out. If total calorie intake is statistically significant than it is a variable influencing obesity - period. I think you are confusing controlling for something and whether a variable influences the DV or not. A variable is normally seen as influencing the DV if it is statistically significant (controlling for other variables). Nothing in your comments above suggests that calorie intake was not significant. It is used for a control on another variable, but that does not mean it is or is not significant. Its p value tells you that.

I am not sure what you mean by "calorie intakes were not controlled" and "Variations in the their calorie intake were only adjusted for in the analysis afterwards." I am not sure how you do that - was calorie intake in the model or not? If it was not, how did they know how to adjust for it?