For more information about PROC MBC and model-based clustering, see Kessler, D. (2019) "Introducing the MBC Procedure for Model-Based Clustering", https://support.sas.com/resources/papers/proceedings19/3016-2019.pdf  ... View more

Use CONTOUR (you have a typo) ... View more

Dup of Solved: How do i get a block diagonal covariance matrix, V, in SAS Proc Mixed ... - SAS Support Communities ... View more

Thanks. I have alerted the manager for that documentation. For the record, the GEE and GENMOD procedures support the EFFECTPLOT statement, but not the EFFECT statement. ... View more

> You can enrich your dataset with new covariates (spline effects) using PROC TRANSREG and subsequently use the procedure of your choice.   An easier choice is PROC GLMSELECT, which supports an OUTDESIGN= option. See the example at https://blogs.sas.com/content/iml/2019/10/16/visualize-regression-splines.html ... View more

I don't know where that sentence came from, but it is not correct. GENMOD does not support the EFFECT statement so the OP must use GLMMIX.   DOC for EFFECT stmt: SAS Help Center: EFFECT Statement says "This section applies to the following procedures: GLIMMIX, GLMSELECT, HPMIXED, LOGISTIC, ORTHOREG, PHREG, PLS, QUANTLIFE, QUANTREG, QUANTSELECT, ROBUSTREG, SURVEYLOGISTIC, and SURVEYREG." ... View more

Look at the R table of parameter estimates. The R model appears to treat all variables as continuous, whereas in SAS you are treating many variables as factors. Resolve that issue first or you'll never get the same estimates.   Also, I suspect you want to add Sex to the CLASS statement so that the Sex variable is treated as a categorical variable.      ... View more

> do I have to have it?  I included that information so that you can easily check the names of the variables in the input matrix. But you can ignore that suggestion if you prefer an alternative.  ... View more

The changes that we've discussed would make your program look like this: proc iml; use pseudo_data; read all var _NUM_ into rc_migr_dist[colname=varNames]; /* Load matrix with movements shares p_ij */ close pseudo_data; nr = nrow(rc_migr_dist); if upcase(varNames)[1] ^= "MK1" then do; rc_migr_dist = j(nr, 1, 0) || rc_migr_dist; varNames = "MK1" || varNames; end; if nr^=ncol(rc_migr_dist) then print "ERROR: Matrix is not square"; rc_migr_dist = rc_migr_dist / 100; /* ??? Is this necessary? */ print rc_migr_dist[r=varNames c=varNames]; use rclassorgcount; read all var {&SOMETHING_THAT.} into rc_orig_count_vec; /* Load vector with number of humans per movement i, Ni */ close rclassorgcount; /* Calc Matrix_normal_u */ M_norm_u = 0; do i = 1 to &K. - 1; maxval = max(abs(i - &K.), abs(i + (&K_Max. - &K.) - 1)); N_i = rc_orig_count_vec[i]; sum_p = sum( rc_migr_dist[i, i+1:nr] ); /*print sum_p;*/ temp = maxval * N_i * sum_p; M_norm_u = M_norm_u + temp; end; create M_norm_u var {"Matrix_normal_u"}; append from M_norm_u; close M_norm_u; ... View more

I suggest you read in the names of the variables dynamically (instead of using a macro) and then add the first column if necessary: proc iml; use pseudo_data; read all var _NUM_ into rc_migr_dist[colname=varNames]; /* Load matrix with movements shares p_ij */ close pseudo_data; nr = nrow(rc_migr_dist); print varNames; if upcase(varNames)[1] ^= "MK1" then do; rc_migr_dist = j(nr, 1, 0) || rc_migr_dist; varNames = "MK1" || varNames; end; print rc_migr_dist[r=varNames c=varNames]; Be aware that this empirical estimate of the transition matrix will be singular. You are specifying a system that will never allow a transition to the first state. If this is not what you want, then use a small positive probability (such as 1E-3) for the first column and adjust the remaining probabilities so that the rows sum to 1. ... View more

If there are 7 states, then each cell indicates the proportion of accounts that transitioned from State=i into State=j during the time period. If no account transitioned to State=7, then the last column of your 7x7 matrix should contain all zeros.   If "add a nullvector" means "add a column of 0s," then I agree. Otherwise, explain a "nullvector" in this context. In matrix theory, a null vector is a vector v such that A*v=0 for the matrix A. But I don't think you are using that definition(?). ... View more

I did a little research, and the OP might be asking about the MWB statistic on p. 24-25 of this paper: Instructions for reporting the validation results of internal models - IRB Pillar I models for credit risk (europa.eu)   However, the paper is analyzing Markov transition matrices whereas the matrix in the OP example is not a transition matrix because the rows do not sum to 1. To learn about transition matrices in IML, see Markov transition matrices in SAS/IML - The DO Loop The probability of reaching a terminal state in a Markov chain - The DO Loop (sas.com)  ... View more

1. What is a MWB? Reference? 2. In your program, the rc_migr_dist matrix is a 7x6 matrix. The error is because you are trying to access the 7th column of a matrix that has only 6 columns. The error is in the loop that iterates over j: do j = i + 1 to &K.; 3. Your matrix is not banded, so I don't understand why your started your post talking about banded matrices. 4. Your matrix does not look like it contains percentages, so I don't understand the comment /* Assuming percentages are given, convert to decimal */. However, if you want to divide the matrix by 100, do it OUTSIDE the loops after you read the matrix:   use pseudo_data; read all var {&rcmigrdist.} into rc_migr_dist; /* Load matrix with movements shares p_ij */ close pseudo_data; rc_migr_dist = rc_migr_dist / 100; /* Assuming percentages are given, convert to decimal */ 5. You can make the error go away by changing the upper limit of the inner loop to &K-1 or ncol(rc_migr_dist) /* No need to loop: sum_p = 0; do j = i + 1 to &K.-1; sum_p = sum_p + rc_migr_dist[i,j]; end; */ sum_p = sum( rc_migr_dist[i, i+1:&K-1] ); 6. I think you would be wise to read the dimensions of the data inside IML by using nrow(rc_migr_dist) and ncol(rc_migr_dist). This is preferable to using a hard-coded macro such as K=7.     ... View more