0

I read somewhere that a large part of BMR is explained by body weight. Using a data in hand, I used step wise regression with BMR as dependent variable and gender, age, height and weight as independent variables and obtained the following models with variables entered at each stage: 

model 1: gender                   R2 = 0.772         
model 2: gender, weight           R2 = 0.979    
model 3: gender, weight, age      R2 = 0.985

Can somebody please interpret what the result I have obtained is?

Andy
  • 18,070
  • 20
  • 77
  • 100
user52672
  • 195
  • 2
  • 7
  • No defensible interpretation is possible using only the information given. At a minimum the number of cases you have is essential. Better would be to use something besides $R^2$ to report the results, such as AIC. Far better than any of those would be to display a scatterplot matrix of gender, weight, age, and BMR (presumably that's "basal metabolic rate"). – whuber Jul 27 '14 at 15:07
  • There are 150 observations belonging to 75 men and 75 women. What is AIC? – user52672 Jul 27 '14 at 15:24
  • See the threads with the [tag:AIC] tag. A top-rated answer appears at http://stats.stackexchange.com/questions/20836/algorithms-for-automatic-model-selection. – whuber Jul 27 '14 at 16:04
  • (i) *stepwise regression* carries dangers; search on it to find numerous posts discussing them; $\ $ (ii) Even without the stepwise issue, models with different numbers of parameters are not directly comparable on the basis of $R^2$... indeed with nested models and continuous r.v.s, the larger model will always have larger $R^2$; $\ $ (iii) but even if you replaced that by something that is more comparable, there's still the danger of trying to compare models on the basis of regression statistics alone -- see [Anscombe's Quartet](http://en.wikipedia.org/wiki/Anscombe%27s_quartet), for example. – Glen_b Jul 28 '14 at 00:50
  • Will a graph of measuerd values against the predicted values help? – user52672 Jul 28 '14 at 11:25

0 Answers0