Re: st: RE: creating combined correlation of dummy (ordered multilevel)

[Stefan Nijssen <stefannijssen@gmail.com>]

Maarten, thanks for your reply. As I am currently reading from your website, the sheaf coefficient seems to be very relevant in my model.

Indeed I am talking about independent variables. The dependent is continuous, a risk factor represented by an interest rate. I am trying to estimate the effect of both a rating scale and a set of accounting ratios to the risk factor. I am hypothesizing that the effect of the rating on the interest rate, compared to the effect of the accounting ratios on this interest rate has diminished in recent years, using data from three points in time during the last decade. Hence the independent variables can be seen as two blocks, the ratings (which are in the form of an ordered scale, AAA (1) AA (2) A (3) BBB (4) BB (5) B (6)). I have split this variable into 6 dummies since the distance between either point might not be the same. Reading from your website, I think I should create this so called sheaf coefficient for both the rating scale and the accounting ratios, and in the ideal case I will be able to visualize the influence of either variable block. Doing this for three points in!
  time, I would be able to see, possibly, a pattern. With the rating scale and the accounting ratios being a possible tradeoff (theoretically), I would be able to see their correlation using the model this way.

Does this sound logical?

Thanks for any help and suggestions,

Stefan Nijssen

On Jun 10, 2011, at 16:48 , Maarten Buis wrote:

> On Fri, Jun 10, 2011 at 3:58 PM, Stefan Nijssen wrote:
>> This is true. The variable AAA AA etc.. is a rating scale, ordered
>> from AAA to B (in original format this was 1 to represent AAA, 2 to
>> represent AA etc. until 6 to represent B, from which through "xi
>> i.Rating, noomit" I created the dummies). I am using the dummies since
>> for instance the distance between AA and A (2 and 3) might not be the
>> same as the distance A and BBB (3 and 4). Therefore I don't think the
>> original variable (with numbers 1 to 6) is the one to read the
>> correlation from, although this would be easiest.
>> 
>> The reason for my interest in the correlation is that theoretically
>> the variable Rating is created using variables I use in the regression
>> in the first place. Any idea how to be able to interpret the
>> multicorrelation?
> 
> This depends on whether you want to use your variable as a
> dependent/explained/y or independent/explanatory/x variable. In the
> latter case I would look at sheaf coefficients to simultaneously
> estimate the distances between the levels and a single effect of your
> variable, see -ssc d sheafcoef- and
> <http://www.maartenbuis.nl/wp/prop.html>. In the former case I would
> look at ordered regression models like -ologit- or -ssc d gologit2-.
> 
> Hope this helps,
> Maarten
> 
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
> 
> 
> http://www.maartenbuis.nl
> --------------------------
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/