Help! my model isn't predicting any yes outcomes and don't know why

E

el89

New Member
#1
#1
I am producing a logistical regression analysis to find a model to predict whether someone has minor mental health problems or not.
I'm almost certain i've coded all my data correctly but no matter which variable I put into the model it doesn't predict any yes outcomes. my classification table of predicted values is always 0 for yes and I don't know why.
I would be very grateful if someone could tell me what i'm doing wrong. it's probably something really simple but i just can't work it out.
Thank you in advance, any help would be much appreciated!
el89 :)
 
Dason

Dason

Ambassador to the humans
#4
#4
Just to make sure... you do have "yes" responses in your data set right? What is the proportion of yes to no?
 
E

el89

New Member
#5
#5
yes there's definitely yes responses. there are 739 yes responses and 3630 no responses in my dataset. i've coded yes as 1 and no as 0. that's right isn't it?
 
E

el89

New Member
#7
#7
yeah thats what i thought. it doesn't work whether i put categorical variables or continuous variables in and i just don't know what else to do :s i really need to get this sorted soon as :(
 
E

el89

New Member
#8
#8
i've tried with other dependent variables too and it's still predicting no yeses. is it my fault or is it a problem with SPSS? have i checked a box i wasn't supposed to or something? it says that it is including all the cases in the model
 
Karabiner

Karabiner

TS Contributor
#9
#9
I'm almost certain i've coded all my data correctly but no matter which variable I put into the model it doesn't predict any yes outcomes. my classification table of predicted values is always 0 for yes and I don't know why.
Click to expand...
You have 739 yes responses and 3630 no responses. Now, if you always predicted "no",
then your rate of correct classifications was 3630/(3630+739) = 83% .

AFAICS the predictive power of your regression model is so low that simply predicting
"no" is still superior to it (maybe you have statistically signficant predictors, which would
not be surprising with such a large sample, but their actual predicitive value is low). Also
have a look at the classification diagram (under "options").

Kind regards

K.
 
You must log in or register to reply here.