logit model without constant: ''no convergence''?

hlsmith

Omega Contributor
#24
I collapsed the extra columns and ran the model using logistic reg with Firth penalization and it was fine. Two of the variables had large SEs but the model converged and generated estimates.


You probably just need to run exact or Firth logistic reg.


I ran it in SAS using the following code:


Code:
[FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][B]data[/B][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2] TS;[/SIZE][/FONT][/SIZE][/FONT]
 [FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]input[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2] c_b    highway    new_investment;[/SIZE][/FONT][/SIZE][/FONT]
 [FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]datalines[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2];[/SIZE][/FONT][/SIZE][/FONT]  
[FONT=Courier New][SIZE=2]  [FONT=Courier New][SIZE=2]   1    1    1
 

1    1    0
 

1    1    2
 

1    1    1
 

1    1    2
 

1    1    1
 

1    2    1
 

1    2    1
 

1    2    1
 

1    2    1
 

0    2    0
 

1    2    1
 

1    2    1
 

1    2    1
 

1    2    0
 

1    2    1
 

1    1    2
 

0    1    2
 

1    1    2
 

1    1    0
 

0    2    0
 

0    2    0
 

0    2    0
 

0    2    0
 

0    0    1
 

0    0    1
 

1    1    1
 

1    1    1
 

0    0    1
 

0    0    1
 

0    0    1
 

1    2    1
 

0    2    0
 

1    1    1
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    2
 

1    1    1
 

1    1    2
 

0    0    1
 

1    0    1
 

0    0    1
 

0    0    1
 

0    0    1
 

0    0    1
 

0    0    1
 

0    0    1
 

0    0    1
  

; [/SIZE][/FONT]

[/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][B]PROC [/B][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][B][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080]LOGISTIC [/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/B][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]DATA[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2] = TS;[/SIZE][/FONT][/SIZE][/FONT]
[FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2]  [/SIZE][/FONT][/SIZE][/FONT] [FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]   CLASS[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2] highway    new_investment    ;[/SIZE][/FONT][/SIZE][/FONT]
[FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2]  [/SIZE][/FONT][/SIZE][/FONT] [FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]   MODEL[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2] c_b ([/SIZE][/FONT][/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]EVENT[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2]=[/SIZE][/FONT][/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#800080][FONT=Courier New][SIZE=2][COLOR=#800080][FONT=Courier New][SIZE=2][COLOR=#800080]"1"[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2]) = highway    new_investment     /[/SIZE][/FONT][/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]      LACKFIT [/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]FIRTH [/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff][FONT=Courier New][SIZE=2][COLOR=#0000ff]CLPARM[/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2]=PL;[/SIZE][/FONT][/SIZE][/FONT]
[FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080] [/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT] [FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][FONT=Courier New][SIZE=2][COLOR=#000080][B]RUN[/B][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][/COLOR][/SIZE][/FONT][FONT=Courier New][SIZE=2][FONT=Courier New][SIZE=2];[/SIZE][/FONT][/SIZE][/FONT]
 
#28
Thanks a lot.
i've been translated and tried to find the most appropriate way to my dataset. the next step was going to be firth logistic regression trying. if you couldnt open, i declare you the estimation results when tried. Are Interpretation for coefficients and odds ratios the same for binary logistic and firth logistic? Its enough if i learn this.
Thanks are not enough, but thanks.
 

hlsmith

Omega Contributor
#29
Estimates and coefficient interpretations are the same as with logistic regression, but you would report that you used the Firth penalization. I will note that when I ran the model the upper bounds on a couple of the odds ratios were very large and were outputted as ">999.999", this is given the same sparsity issue that was brought up before.


I will explicitly add on this, the issue is that you have an outcome with one group having only 20 reported events and your model has two variables with three categories meaning you have nine different subgroups and two outcomes, so you have 18 possible combinations, but only 53 observations. It is difficult to get conclusive statistical results with such small subsample sizes.
 
#30
Thats exactly right. I have small sample and trying to do the best -right, economically logical and econometrically significant- method for my data.
The more explanatory variable, the better study,, somebody embed this in our heads. but wrong.
Following your words, I'll try to miniaturize (even two responses each,if required) IVs. Apply Firth regression (i've never heard before). Estimate coefficients, odds', goodness of fit tests. Then contact you again. I hope you/it would be the next stop for my analysis.
Great of you. Best wishes. Good work.
 
#32
My logistic model's dependent variable is cost/benefit ratio of an investment. It takes value ''1'' if c/b ratio>=1 and ''0'' if ratio<1.
If you categorize like this you will lose a lot of information. It is better to use the original data. (Original cost/benefit) Or to use the difference benefit - cost. (If "benefit" is the same as revenues, it is customary to take revenue - cost = profit.)

But I believe that the main problem in this thread was with the iv:s.
 
#33
Dear Greta. Your opinions about using original data are so sensible.
But I got dependent as ratio; i -accurately- have no all revenues or expenditures (costs).
First tried original data but fully insignificant and not interpretable results. than i fit logistic regression with unrefined IV:s; (like 0-1-2 responses or not very well designed status) and i saw good results than my data was explained better by this way.
But then ''not concave'' incubus began.
I hope you read previous posts, see my whole dataset and agreed. Im recommended to try Firth regression. ANd i'm very happy as you think the main problem is with the independents.
As you and hlsmith said, i'll reorganize Iv:s, try to make them smaller.
Thanks for interest. Im going to do the best thanks to you.
 
#34
sorry but adding something else. im getting endless iterations and im typing ''break'' option by hand, so stata gives r(1) error code because of breaking.
But the convergence problem gives r(430) code in stata and estimating the model coefficients (but not st.errors) despite non-prodiving the convergence.
page 8 is the place looks like my ''not concave'' error, but not the same: https://www.stata.com/manuals13/rlogit.pdf
So, all the clues point that there's a convergence problem but i cannot prove by this way.
Sorry for taking time but im really ambitious about doing the correct one.
 
#37
Respectable Greta.
Dataset posted #19,20,21,25,26.
None of these contains the original data (of the dv of b/c) before the recoding to 0/1. As I said, you lose information (get less efficient estimates) and get problem with convergence.

But, never mind. Good luck!
 
#38
How can i not mind? i just didnt understand you wanted the original c/b's. I have to pick them up again one by one from each project before sending you.
Thanks for interest. I'll do this and come back.
 
#39
I have tried thousands of combinations as all you both said before: with all variables, limited number of variables, one by one, 2 or more responses, etc.
Odds ratios always been the problem of all models.
Here are the best results.
Please remark your ideas about the models
PS::: The reason i didnt apply regression before is constitutional. The constitute gave me the c/b ratios in return to i guarantee not declare them. So i classify them as 0 or 1 so as not to show the ratios. I hope i could explain..
thanks for all..