I am doing research related to efficient fund utilization of Swachh Bharat Mission by different states of India. I have collected data related to institution, demography, ease of doing business, etc. I have calculated relative efficiencies of Indian states towards utilizing the funds for Swach Bharat Mission and delivering tangible outcomes. I have used the technique of Deta Envelopment Analysis for the same. Kindly help me with the issue. I have tried bootstrapping, but it did not work well.
The short answer is you don't. You don't say anything about the DV so I'll assume we're talking OLS regression. other choices are somewhat different. I'll tell you a rule of thumb for OLS: A min of 5 obs/IV. You are nowhere near that so standard OLS is out. I am going to assume that you want to do prediction-you don't say but you have to get rid of some IVs. I will suggest adaptive lasso variable selection.. See the stuff at this link including the papers esp concerning the oracle property and the R program at the very end which you can use:
https://www4.stat.ncsu.edu/~boos/var.select/lasso.adaptive.html
. Here are the reasons:.
1 lasso runs even if p>n
2. The oracle property says that the variables selected are the optimal predictor set of you candidate IVs..- solves this problem
3. This final model is a predictive model not explanatory (see the Shmueli reference in our papers for full details). All of these references should help. Please ask again if you have further questions. Best wishes, David Booth BTW here is a source to get you started in R. I would look at this link: http://sgpwe.izt.uam.mx/files/users/uami/gma/R_for_dummies.pdf
Abhishek Wagh
Your number of observation (or sample size) is less than 30. In this way, the regression estimate will not provide you accurate estimate. My suggestion is to increase to sample size greater than 30 or larger.
If it is not possible for you to increase the sample size, then you may estimate your regression by using structural equation modelling through SmartPls software as it supports smaller samples also.
The short answer is you don't. You don't say anything about the DV so I'll assume we're talking OLS regression. other choices are somewhat different. I'll tell you a rule of thumb for OLS: A min of 5 obs/IV. You are nowhere near that so standard OLS is out. I am going to assume that you want to do prediction-you don't say but you have to get rid of some IVs. I will suggest adaptive lasso variable selection.. See the stuff at this link including the papers esp concerning the oracle property and the R program at the very end which you can use:
https://www4.stat.ncsu.edu/~boos/var.select/lasso.adaptive.html
. Here are the reasons:.
1 lasso runs even if p>n
2. The oracle property says that the variables selected are the optimal predictor set of you candidate IVs..- solves this problem
3. This final model is a predictive model not explanatory (see the Shmueli reference in our papers for full details). All of these references should help. Please ask again if you have further questions. Best wishes, David Booth BTW here is a source to get you started in R. I would look at this link: http://sgpwe.izt.uam.mx/files/users/uami/gma/R_for_dummies.pdf
Agree with the above. You should consider data reduction on your DVs - limited to 2 or 3 orthogonal measures.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size