Thank you so much for a response well more informative than I originally anticipated! Have a nice weekend! ... View more

I used to create plots where different groups are displayed in different color and line patterns and therefore encountered the AttrPriority option in ODS GRAPHICS. I read the relevant introduction to that option given in SAS Help, but I was still rather confused on its utility. Could you please briefly explain it? Thank you! ... View more

I think some data post-processing needs to be conducted to get what you want. Now that SAS has obtained the cumulative hazard function levels at different time points, you can subtract the cumulative hazard function right before a given time from the one at the given time to obtain the estimated baseline hazard function of that time. In formulae, suppose we are interested in estimating h0(t1) and t0 is the latest time point of an event right before t1. Let the cumulative hazard function estimators output by SAS be denoted as H(.)_hat. Then, what you get from SAS is H(t1)_hat and H(t0)_hat. Use the XBETA= option in the BASELINE statement to get the linear combinations of predictors. Then, you can get H0(t1)_hat=H(t1)_hat/exp(XBETA) and H0(t0)_hat=H(t0)_hat/exp(XBETA). Therefore, h0(t1)_hat=H0(t1)-H0(t0). This method works fine in the absence of ties. But, when there are, which is a common scenario, then I think directly computing the Breslow estimator of the baseline hazard function can be more helpful. In the formula, di is an indicator variable of whether an observation experienced an event at ti. R(ti) is the risk set of ti. In other words, the denominator is the sum of linear combinations of predictors (i.e., XBETA's) of all subjects in the risk set of ti. ... View more

For the first question, I think only the writer of the relevant documentation can give you a definite answer. It seems to me that the Weibull formula in the SAS documentation is wrong. I think w_iy_i's should replace the y_i's in the formula. However, an easy way to see if the WEIGHT statement in PROC FMM works properly is to build one Weibull model with the WEIGHT statement and build another one without that statement and compare the results. If you wish to have a definite answer as to whether PROC FMM maximized the log-likelihood function in exactly the same way you wanted it to be, then a more complicated approach is to compose custom log-likelihood function in SAS procedures like PROC NLMIXED, maximize the log-likelihood there and compare its result to the one generated by PROC FMM. For the second question, I think whether a WEIGHT statement is specified is far more important than what the specified weights stand for. In other words, I think PROC FMM will give you the correct result as long as you have stored the weights in a variable and specified it in the WEIGHT statement, regardless of what the weights stand for and what they are supposed to do. Again, to see if PROC FMM gives you exactly what you wanted, you can also custom your log-likelihood function and maximize it in procedures like PROC NLMIXED and compare its result to the one obtained by PROC FMM. ... View more

@David_M wrote: In general, can you explain to me why my simple survey a complex survey for SAS? I only have two groups of people with only 3 observations. First of all, whether a survey is a complex survey has little to do with how many observations a group have, whether a survey can be called "complex" is determined by its sampling scheme. Those with multi-stage sampling, stratification and clustering in the sampling scheme are called complex surveys. By the way, didn't you say in your last post that your dataset contained 400 observations? Why is it 3*2=6 now? Second, whether a survey is complex itself matters little as statistical methods for complex surveys are readily suitable for surveys of all kinds of complexity. Therefore, whether your data is survey data is important, whether or not your survey data is complex is not important in terms of making decisions on the statistical methods to use. ... View more

@PaigeMiller wrote: I don't know what handling categorical values with missing values has to do with this thread. Take a look at the abstract of my citation before you post your response, Paige. I do not post meaningless citations and I do not think it responsible for you to make such remarks. @PaigeMiller wrote: Clearly, PLS has been modified to include categorical predictors, as SAS has been able to do it for at least 20 years now. I don't know if SAS gets the credit for this invention, or if it had been published previously elsewhere. Well, then answer my question. Does the generic PLS algorithm have a mechanism that restircts certain independent variables to only take certain values when the variable contributes to a factor? For instance, if I have a binary indicator for death vs. alive that take 1 when a person is alive in my predictors, what does β*0.5 mean? Is it meaningful? Can PROC PLS prevent the formation of such meaningless elements in the factors generated? ... View more

@David_M wrote: My apologies for not specifying at the outset that I was dealing with survey data. I didn't think it mattered if my confounders were survey based or not. Never mind. But please be clear from now on that complex survey data complicates everything concerning statistics. Every statistical method you employ have to be complex survey data-adapted ones, not the generic ones. In fact, I am not aware of any deterimental effect collinearity would have upon the parameter estimates of the regression coefficients in logistic regression, but I know that they are severly biased if you neglect the complex survey nature of your data. It would therefore be much better if you pointed out that your data originated from complex surveys in the first place the next time you raise a question on complex survey data analysis. @David_M wrote: Thank you for your suggestions. Yes, this is a matched paired survey from a longitudinal study of 400 healthcare workers whose characteristics I'm trying to analyze at two time points only, pre- and post-Covid. The characteristics are Job Satisfaction (JS), Intention to Leave the workplace (ITL) and Respect/Civility at Work (RW), which are in the form of Likert scales (1 - 4 and 1 - 5). I want to develop 2 models by performing a standard logistic regression between JS and RW and another between ITL and RW at both pre and post-Covid time points, so 4 models total but similar analytical methods for all 4 models. The results, conclusions and implications for each model would be different.   As previously mentioned, I've identified 30+ mixed type categorical and continuous confounders that I want to add to my models but I need to eliminate or drastically reduce any correlation between them, hence preventing unstable regression estimates, inflated standard errors, etc. These confounders will be treated as standard predictors in my models. All 4 models will have and end up with these same set of de-correlated confounders.   Are there no SAS procedures or methods suitable for this task? Unfortunately, to the best of my knowledge, the answer is "yes", unless you learn the formulae and compile codes on your own in SAS. By the way, complex survey data analysis is a field of statistics receiving not that much attention, so there are really few researchers working on how to conduct statistical diagnostics like detecting for collinearity and correcting them. In fact, as Taylor H. Lewis, author of the book Complex Survey Data Analysis with SAS | Taylor H. Lewis | Taylor & Francis noted, statistical diagnostics of models for complex survey data are "in its nascent stage". It is therefore highly likely that you encounter a particular problem in this field and discover that to date, there is no way of dealing with it. ... View more

OK, so now that your problem concerns survey data analysis, things are much more complicated than I anticipate. We all did not know that your data involved survey data analysis, the appearance of which challenges the plausibility of all our previous suggestions. Anyway, there has been R packages designed for testing for collinearity in complex survey data analysis, like CRAN: Package svydiags and svycollinear: Condition indexes and variance decompositions in general... in svydiags: Regression Model Diagnostics for Survey Data. On the other hand, methods like Variable selection with LASSO regression for complex survey data - Iparragirre - 2023 - Stat - Wiley Online Library could be used for dealing with collinearity. ... View more

I am not an expert on deep learning, so I thought I should not respond to your question before. But now that three weeks have passed since you raised your question and no response have been offered, I would like to express my concern so that you may find a better place to ask your question. The concern is: are you sure you are dealing with SAS codes? I do not have access to SAS Viya, the software developed by SAS Institute capable of doing deep learning, but the formats of your code really makes me think that your code is highly unlikely to be a SAS code. For instance, SAS does not use hashes to mark the beginning of appendices; SAS does not have codes like "model.add" (with dots in the middle); SAS does not stuff multiple options in a space flanked by parentheses. So I really speculate that you are coming to the wrong forum for help. The users here are all SAS users. While being a SAS user never implies that he/she cannot use other softwares, the fact that your question has been here for three weeks and no one has ever responded urges me to issue my concern. Nobody in this forum might be able to answer your question because you might had raised it in the wrong place. Therefore, if your code is not a SAS code, then confirm the software on which the code should run and post your question on a forum on that software to minimize your time to get an answer; if I am wrong and your code really is a SAS code, then my response brings you back to the first place of the Statistical Procedures board so that other people visiting this board can find it more easily. Good luck! ... View more

Yes (It seems that Paige has not responded to your question. I am not sure of the underlying reason, but I will respond your question anyway in case you wait too long). By the way, the message I would like to add is that PROC PLS is suitable for building PLS linear models. In other words, the dependent variables of the model have to be continuous if you wish to apply PROC PLS. On the other hand, you cannot build PLS logistic regression models with PROC PLS. Other software packages have to be employed in this case. ... View more

Thanks for the feedback. But to the best of my knowledge, the "generic" PLS algorithm treats categorical independent variables the same way as it treats continuous ones. In other words, the fact that categorical variables only take finite values is not taken into consideration in the "generic" PLS algorithm. I am not sure the extent to which adapted PLS algorithms like the one I cited in my previous reply to @David_M take this into account as I have not stuided on PLS in the recent years. ... View more

@David_M wrote: Thank you for the fast response. Based on the article you posted and some other reading, yes, PLS is great de-correlator when used with highly collinear variables. But what I've read so far, all these variables are continuous only. Am I correct in this?   The independent variables of PLS are chiefly continuous ones. To me, that is a major limitation of PLS. To the best of my knowledge, there have been variants of PLS that can accomodate categorical independent variables as well (e.g., Treatment of Categorical Variables with Missing Values Using PLS Regression | SpringerLink). Another way to deal with this is to use penalized regression like ridge regression and LASSO. ... View more

Take a look at this discussion, which is a question I raised about two years ago in this forum: Solved: How can I perform principal component analysis for logistic regression... - SAS Support Communities. A reference I find extremely helpful is a SAS usage note named 32471 - Testing assumptions in logit, probit, Poisson and other generalized linear models. Special thanks are given to @StatDave for introducing me this note. By the way, I agree with @PaigeMiller's remarks on collinearity: you cannot eliminate collinearity, you can only somehow manage to alleviate its impact on results. ... View more

Unfortunately, I do not know. Maybe you can refer to the formula itself and try to compile codes on your own. ... View more

Take a look at this article: Sample size calculations for cluster randomised controlled trials with a fixed number of clusters | BMC Medical Research Methodology | Full Text. An R package is also available: CRAN: Package CRTSize. In addition, a book on this topic is available: Sample Size Calculations for Clustered and Longitudinal Outcomes in Clinical Research. ... View more