s**********n
发帖数: 111 | 1 Question: which model is better?
Model_A: logit(p)=X1+X2+X3+X4+X5+X6+X7+X8+X9
Assume contribution of each variable is:
11%, 11%, 11%, 11%, 12%, 11%, 11%, 11%, 11%
Model_B: logit(p)=X1+X2+X3+X4+X5
Assume contribution of each variable is:
15%, 15%, 15%, 10%, 45%
Thank you a thousand for the help!!! | n**m
发帖数: 156 | 2 Variable contribution in explaining variation is not a model selection
criteria. You should look at AIC, R-square kind of stuff. just my 2 cents. | d******o
发帖数: 59 | 3 if R^2 is the same. consider BIC. the second one should be better. | a******n
发帖数: 11246 | 4 先看各种指标比如R-squared, adj. R-squared, mellow's Cp, AIC, BIC,...
反正公说公有理，婆说婆有理。
然后要考虑p的范围，比如有些model p是集中在0或者1附近的，
这种情况下可以看ROC curve，比较sensitivity神马的。
希望大牛们指正+补充。
【在 s**********n 的大作中提到】
: Question: which model is better?
: Model_A: logit(p)=X1+X2+X3+X4+X5+X6+X7+X8+X9
: Assume contribution of each variable is:
: 11%, 11%, 11%, 11%, 12%, 11%, 11%, 11%, 11%
: Model_B: logit(p)=X1+X2+X3+X4+X5
: Assume contribution of each variable is:
: 15%, 15%, 15%, 10%, 45%
: Thank you a thousand for the help!!!
| c****s
发帖数: 395 | 5 I am not good at this, but here is my 2 cents
usually contribution is not used in selecting models
but here comparing two models
the second one, the last variable has a comparatively huge contribution.
it means if it change a little, the model result will change greatly.
if there are a little more outliers for this variable, then the model result
will far from correct in predicting.
so based on that, I will choose model 1.
【在 s**********n 的大作中提到】
: Question: which model is better?
: Model_A: logit(p)=X1+X2+X3+X4+X5+X6+X7+X8+X9
: Assume contribution of each variable is:
: 11%, 11%, 11%, 11%, 12%, 11%, 11%, 11%, 11%
: Model_B: logit(p)=X1+X2+X3+X4+X5
: Assume contribution of each variable is:
: 15%, 15%, 15%, 10%, 45%
: Thank you a thousand for the help!!!
|
|
