Yes, we converted some equations to latex that were images. But one has to pay for it.

]]>Sent from my iPhone

>

]]>I enjoyed reading your article but I do have a few questions:

* When playing around a bit with your selection criterion I noticed that quite often a 3rd or 4th order polynomial was prefered when I generated my random data based on a 2nd order polynomial because the variance for those orders was a slightly lower than for the 2nd order polynomial. In your article you mention that once there is no significant improvement anymore you’ve found your optimal order of your polynomial which I understand but find difficult to implement given that it is quite vague. Nonetheless, I presume we could come up with a rule of thumb to define ‘significant’. However, that way the criterion becomes more and more subjective in my opinion. What’s your take on this?

* My first thought to tackle the same problem was to simply use the widely known BIC criterion. When playing around with some dummy data this seemed to be a bit more robust. Apart from that it is also less subjective given that no additional definition for ‘significant’ improvement is required. Do you see any shortcoming in using BIC?

Thanks a lot for your article!

Arno

]]>I enjoyed reading your article but I do have a few questions:

* When playing around a bit with your selection criterion I noticed that quite often a 3rd or 4th order polynomial was prefered when I generated my random data based on a 2nd order polynomial because the variance for those orders was a slightly lower than for the 2nd order polynomial. In your article you mention that once there is no significant improvement anymore you’ve found your optimal order of your polynomial which I understand but find difficult to implement given that it is quite vague. Nonetheless, I presume we could come up with a rule of thumb to define ‘significant’. However, that way the criterion becomes more and more subjective in my opinion. What’s your take on this?

* My first thought to tackle the same problem was to simply use the widely known BIC criterion. When playing around with some dummy data this seemed to be a bit more robust. Apart from that it is also less subjective given that no additional definition for ‘significant’ improvement is required. Do you see any shortcoming in using BIC?

Thanks a lot for your article!

Arno

]]>