topic Re: Factor scores for Reduced Rank Regression in New SAS User

Factor scores for Reduced Rank Regression

JianJian — Fri, 11 Nov 2022 14:44:46 GMT

Hello, I am using RRR for an analysis of Dietary pattern. I have data for two sets of data for 2021 and 2022, where data for the response variable is missing in 2021, so I want to calculate the 2021 meal score using the dietary pattern coefficients generated by the 2022 RRR. So far I have extracted Xloading, how to calculate the final meal score for each sample. Thanks!

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Fri, 11 Nov 2022 15:32:59 GMT

Include the 2021 data with the missing response in your data set that also has the 2022 data, and then run RRR and ask for the output to include scores and predicted values.

More details here: https://blogs.sas.com/content/iml/2014/02/17/the-missing-value-trick-for-scoring-a-regression-model.html

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 03:12:22 GMT

I'm very sorry, because I am a new user to SAS, I am not familiar with it, the following is my code for RRR, how to do predict I still can't, can you please tell me how to write the later predict code. Thank you!

proc pls data=data2 method=rrr varss nfac=2 details censcale;
model e49 e11=
f1 f2 f3 f4 f5 f6 f7 f8 f9 f11 f12 f13 f14 f15 f16 f17 f18 f19 f20 f21 f22 f23 f24 f25 f26 f27/solution;
output out=data3 XSCORE=xscore YSCORE=yscore;
run;

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 10:49:32 GMT

The SAS data set used in your code which you named data2 should contain both the 2021 and 2022 data. You need to modify the OUTPUT statement in PROC PLS to include the PREDICTED option.

output out=data3 XSCORE=xscore YSCORE=yscore predicted=pred_e49 pred_e11;

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 11:21:26 GMT

I'm really sorry, factor score of dietary pattern = factor loadings * standard food intakes, why should I predict the dependent variable?

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 11:33:06 GMT

@JianJian wrote:

I'm really sorry, factor score of dietary pattern = factor loadings * standard food intakes, why should I predict the dependent variable?

I don't understand the question.

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 11:49:43 GMT

I'm very sorry, I didn't make my question clear, my dependent variable is now only available in 2022, but I have a diet for two years, should I first run with the data of 2022, then predict the standard dietary intake in 2021, and finally calculate the factor score, at present I will only do the RRR in 2022, but will not derive the factor score of the entire population

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 12:55:40 GMT

I have already explained this. The data used in PROC PLS should contain BOTH 2021 and 2022. The scores and predicted values will be computed for all of the data, 2021 and 2022, and the model fit will be based on just 2022 data (since the 2021 data has no Y values).

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 15:58:29 GMT

ADDING: Since you have 27 x-variables and 2 y-variables, I would think PLS would be a better choice than RRR for this data. It will handle the multicollinearity of the X variables better than RRR will.

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 16:12:10 GMT

谢谢你之前的回答。现在，我又有一个问题，为什么基于 xweights 和标准化食物摄入量的产品总和不等于直接从 proc pls 获得的因子分数（输出 xscore=）？

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 16:16:53 GMT

Translation: "Thanks for your previous answer. Now, I have another question, why the sum of products based on xweights and normalized food intake is not equal to the factor score obtained directly from proc pls (output xscore=)?"

Show the code you used. Show output that illustrates the problem.

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 16:36:12 GMT

Data2 包含 2021 年和 2022 年。我使用的代码如下，proc pls data=data2 method=rrr varss nfac=2 details censcale;型号 E49 E11= F1 F2 F3 F4 F5 F6 F7 F8 F9 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27/解决方案;输出输出=数据3 XSCORE=xscore YSCORE=yscore predicted=pred_e49 pred_e11;跑;

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sat, 12 Nov 2022 16:39:15 GMT

I said: "Show the code you used. Show output that illustrates the problem."

I don't see any output that illustrates the problem. You need to provide enough detail so I can look at it and see the problem.

Re: Factor scores for Reduced Rank Regression

JianJian — Sat, 12 Nov 2022 16:58:54 GMT

proc pls data=data2 method=rrr varss nfac=2 details censcale;
型号 E49 E11=
F1 F2 F3 F4 F5 F6 F7 F8 F9 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26 F27/解决方案;
输出输出=数据3 XSCORE=xscore YSCORE=yscore predicted=pred_e49 pred_e11;;
跑;

我比较了 stata 计算的饮食评分和 SAS 计算的 xscore 没有一致性，但文献说 SAS 和 stata 的评分应该是一致的

Re: Factor scores for Reduced Rank Regression

PaigeMiller — Sun, 13 Nov 2022 11:02:29 GMT

What "literature" says the scores from SAS and STATA should be the same? In multivariate procedures such as this, different algorithms can produce different results, which can be entirely different (which may or may not be a problem), or the scores can have the opposite sign (which is not a problem), or they can produce the same results but scaled differently (which is not a problem either).

I will not download attachments. If you are going to show us SAS output, you need to include it in your reply by either copying and pasting as text, or as a screen capture using the "Insert Photos" icon.