search

tom-hc-park / STAT550-450-for-Seniorworkers-from-Korea

0 stars 0 forks source link

New results of forward Adjusted R^2 #16

Open liuzhen529 opened 6 years ago

liuzhen529 commented 6 years ago

Hi Hyunok,

I did the stepwise Adjusted R^2(forward). The process is similar to forward AIC, which is adding variables one by one from null until the adjusted R^2 does not increase.

The results are(the order corresponds to the order they are picked): 1)numeracy: GENDER, AGE, ED_Level, pub_priv, work_flexM, act_lrn, FNFAET12NJR, NFEHRS(adj R^2 = 0.35131)

2)literacy: GENDER, AGE, ED_Level, pub_priv, work_flexM, act_lrn, NFEHRS( adj R^2 =0.29872)

The results are very similar to the stepwise AIC. "FNFAET12NJR, NFEHRS" are very tricky one. Though adding them can increase the adjusted R^2, their p-values in the model are larger than 0.05 and more variables are included.

1) For numeracy, the adj R^2 for "GENDER, AGE, ED_Level, pub_priv, work_flexM, act_lrn" is 0.35020, which is just 0.0011 smaller than the previous one. Also, after using ANOVA Comparison, "FNFAET12NJR, NFEHRS" cannot add significant information.

2) For literacy, the adj R^2 for "GENDER, AGE, ED_Level, pub_priv, work_flexM, act_lrn" is 0.29794 which is just 0.0008 smaller than the previous one. Also, after using ANOVA Comparison, "NFEHRS" cannot add significant information.

If you want to use adj R^2 as the only criteria, it is okay and reasonable. But due to parsimony, I would suggest using simpler and sufficient model for both dependent variables: "GENDER, AGE, ED_Level, pub_priv, work_flexM, act_lrn".

Let me know if you have further questions.

Sincerely, Zhen

hyunokryu commented 6 years ago

I appreciate this. I will report it and get back to you. Thank you!

hyunokryu commented 6 years ago

Hi, the professor I work with requested the results promptly and so I ran SPSS to analyse the data with all four dependent variables. it would be interesting to compare the results later with what you will come up with. I have the summary of the results (with R^2 values) with me I can share them with you if you are interested.

In the last meeting, the team members mentioned about "predictive power" I wonder if you can run the analysis so that we have the prediction values too?