EViews.com

Hi,
I am working with Eviews 7.

I try to run an ordered model with 4 independent variables but I get the error message: Insufficient number of observations. However if I run the model with only 3 independent variables ( It doesn't matter which specific variable I drop) it works perfectly.

Is this a bug in Eviews or can I fix this problem by myself? ( I remember I had some troubles with this as well in Eviews 6, but after downloading the latest patch it was fixed)

Thank you! Kind regards

Hard to tell without seeing the workfile. It could be that the 4th variable has lots of NAs, which would reduce the number of observations, which could cause the error message to be valid.

No this isn't the case. None of my variables has NA. Additionally I tried it with different dependent variables. Exactly the same number of observations and exactly the same independent variables. In that case it works perfect. I just can't find the difference in the data why it won't work. The only thing I can come up with is that the variance of this specific dependent variable is bigger.

Can I show my workfile?

Sure, either attach it to a post here, or email it to us at support@eviews.com

Well, I've had a look, and what's going on is that you have an "Insufficient number of observations" :)

The difference between your two specifications is that the latter dependent variable (INTEGER_TOTAL) has more distinct categories than the former (INTEGER_DOWN): 70 versus 61. As you know, the ordered discrete choice model estimates a separate coefficient for the cutpoints between each category so you'll have 69+number of variables coefficients to estimate in the former case, and 60+number of variables in the latter. With 72 available observations, the 69+4 leads to the insufficient number of observations error, while the 60+4 does not.

I will point out that you probably don't have enough observations to reasonably estimate the INTEGER_DOWN model. The nice thing about an ordered discrete choice model is that it only uses the order not the cardinal values of your dependent variable, so you are not making parametric assumptions about a functional form for the cutpoints. The downside is that because of this, there are no smoothness conditions placed on the cutpoints, hence the need to estimate a separate parameter for each. But 72 observations really isn't enough to estimate this model, since the asymptotics require that the number of observations for each category go to infinity.

The bottom line is that the error message is an accurate one and that you may want to think about whether this is the best specification for you, given the data constraints.

I hope this helps.

Hi Glenn,

Thanks a lot for your quick and clear answer. I didn't look at the problem in this way.

Unfortenately I can't obtain more data, so this model is not accurate for my data. Regarding the characteristics (ordinal) and the size of the data no other models will be accurate as well, right?

Thanks for the new insights anyway!

You could always group observations into a few categories, say a LOW, MED, HIGH, or some other such grouping of the ordinals. That would help.

Thank you so much Glenn! Problem solved :D