[讨论]SPSS CATREG Procedure Interpretation

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

I am running CATREG for hypotheses testing asnd found significant relationships among my variables. I need to identify the direction of the relationship (such as which parenting style -nominal- results in higher peer
status -ordinal). Since coefficients table reports standardized beta values I can only see the degree of relationship not the direction. For example Itook one of those variables (parenting style) and entered into analysis as
is and created dummy variables for categories as well for second run. Neither way provided the detailed results I need. How can I get direction of relationship from CATREG?

Thanks in advance.

David

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：Procedure Interpret CATREG interp inter 讨论 SPSS Procedure CATREG

Look at the quantification tables or transformation plots. For example, if parenting style has categories 1,2,3,4 with quantifications -2, 1, -1, 2 and the beta is positive, then categories 2 and 4 are related to the categories with the higher quantified values of the dependent
variable and categories 1 and 3 to the categories with the lower quantified values.

If dependent variable ordinal scaling level, then the higher quantified values always correspond to the categories with higher category numbers and lower quantified values to categories with lower category numbers. But with nominal scaling level you have to look at quantification table or transformation plot to see which categories correspond to higher/lower quantified values.Is this what you are looking for?

Anita van der Kooij
Data Theory Group
Leiden University

Each category dummy measures the effect of being in that category, relative
to the reference category (the omitted category that obtains when all other
dummies are zero). If the effect for one category is -0.5, you know the
effect is to reduce the value of the dependent variable (relative to the
reference category). Therefore the sign of the coefficient gives you the
direction (slope) of the relationship. If by "direction" you mean causal
direction, i.e. whether A causes B or the other way around, that choice is
made when you choose to use one variable as the dependent variable and
others as independent variables (which you may regard as predictors or
causes). If the various categories of a variable have no intrinsic order,
these categories may have effects going up or down in whatever order. You
may choose to reorder the categories going from the highest to the lowest
coefficient, or the other way around, with the proviso that those
coefficients are only valid within your regression equation (i.e.
controlling for your other variables in the equation); redoing the
regression with other control variables may change the coefficients and
their relative order.
Hoped this helps.
Hector

I benefited from the comments by Anita did as I was advised. It worked fine for my other variables but the one I used as an example in my message acts weirdly. And your comments Hector, helped me to realize the identification problem for categories (at least in my case)

Parenting style has three categories: Authoritative, Permissive and Authoritarian.
In my dataset first one yields more positive results than others in other forms of analysis (linear regression with dummies etc.)
I applied Crosstabs and Chi-square tests confirmed this relationship.

However when I entered this variable into CATREG as integers (Authoritative = 1, Permissive = 2 and Authoritarian =3) I receive following results as quantifications. This is opposite of what it should be. If variables are recoded in reverse order sign of the quantification changes. I select nominal scale with no discretization for this data.

Parenting Style_PSDQ(a)

Category
Frequency
Quantification
Authoritative
271
-1,289
Permissive
310
,411
Authoritarian
186
1,193

a Optimal Scaling Level: Nominal.

If I recode this variable into string format and enter it into analysis "ranking" is selected as discretization automatically and I receive following results which make sense for the dataset.

str_parent(a)

Category
Frequency
Quantification
AT
186
-1,193
AV
271
1,289
PM
310
-,411

a Optimal Scaling Level: Nominal.

(I didn't bother with full labels in my trial :-)

It is same for dummy variables too Hector.

After my trials upon your advices my question becomes what parameters should I use in my CATREG and what should I do for a nominal variable to get unbiased results? As in parenting style the result changes if I change arbitrary order of numerical identification of the categories. Should I take the cumbersome work and convert all nominal variables used in CATREG to string or is there any other way to have CATREG treat this variables equally regardless of what I use to identify them.

Thanks for the help again

David

So, your results with coding categories one way are:

Category Frequency Quantification
(1) Authoritative 271 1,289
(2) Permissive 310 , 411
(3) Authoritarian 186 1,193


And with coding another way:

(1) AT 186 -1,193
(2) AV 271 1,289
(3) PM 310 -,411



The result are the same apart from sign change, then the beta coefficient also had a sign change(right?). So, the contribution of the variable to the predicted value is the same. For a nominal variable the coding is irrelevant, so you are free to choose for the coding that results in the sign that you prefer.



note: if you replace/recode the categories of nominal variable with the B coefficients (not the beta's), resulting from linear regression with dummies, and center this variable and normalize on N, the resulting values are equal to the CATREG quantifications.



Regards,

Anita

The note in previous mail was not complete



note: if you replace/recode the categories of nominal variable with the B coefficients (not the beta's), resulting from linear regression with dummies, and center this variable and normalize on N, the resulting values are equal to the CATREG quantifications.



I should have added that you have to replace the category omitted in linear regression with dummies with a zero to obtain the CATREG quantifications.



Anita

Thanks for the comments.

I tried it with dummy variables setting them as ordinals with ranking discretization and obtained similar relationships with Crostabs/Chi-square analysis.

But I am still not clear in one point about CATREG. If I disregard the signs, how can I find which one of the categories increases or decreases peer status without invving dummy variables? I a m in an expectation that CATREG eliminates the need for dummy variables and provide similar information through quantifications. If I am not mistaken isn't it a contradiction?

>But I am still not clear in one point about CATREG. If I disregard the signs
You should not disregard the signs, but look at the sign of both the quantifications and the beta coefficient.

>I a m in an expectation that CATREG eliminates the need for dummy variables and provide similar information through quantifications.
Right.

>If I am not mistaken isn't it a contradiction?

? is what a contradiction?

What I mean by contradiction is that the analysis supposed to simplfy regression with categorical variables by avoiding need for dummies but when I change the arbitrary numbers represinting nominal categories results change. that was the point confused me.

Now I have a question about checking both Beta and Quantifications. If beta is negative and category #1 has negative quantification while #2 and #3 has positive and 3 is higher, what would we conclude from this situation?
FYI: Chi-square analysis in crosstabs function reveals cateagories are ordered as 1 -2 -3 if we order them according to their positive impact on dependent variable.

David Clemson <davidclemson@gmail.com>

Thank you very much Anita. After some reading based on your comments I am pretty much done with this matter.

David