About amyip

amyip

Thank you for the thoughtful explanation

amyip

Thank you so much -- this code worked perfectly! so much more elegant than 14 if-then-else statements.

amyip

Very helpful to know about DO loop ending with value of 2023 -- I did see that in the log.

amyip

Those were typos from retyping quickly. Should have been: array years[2009:2002] Y2009-Y2022; Thanks for your feedback about the brackets -- I did find that confusing in the articles/blogs I was reading.

amyip

Yes, absolutely typos -- using SAS EG on a secure server so can't copy and paste, have to retype everything. Should have been: array years[2009:2002] Y2009-Y2022; Thanks for the feedback

amyip

I'm using proc nlmixed which I understand doesn't have a class statement, so I need to manually create dummy variables. How would I use an array to create dummies Y2009-Y2022 for existing variable "year" with values 2009-2022? I tried adapting code on p. 9 <https://support.sas.com/resources/papers/97529_Using_Arrays_in_SAS_Programming.pdf>, but it's not working: array years[009:2002] Y2009-2022; do year=2009 to 2022; years(year)=0; end; years(year)=1; Thanks for any feedback!

amyip · ‎08-29-2024

In follow-up to proc nlmixed vs. proc fmm, I was able to run my model with both procedures, but they yielded nonidentical (albeit similar) results. Is one more accurate than the other? From this note: Although compared to proc nlmixed, proc fmm provides a much easier way to specify a zero-truncated poisson regression, it has much more limited postestimation options. For example, estimate and predict statements are not available with proc fmm. Additionally, proc fmm like most other SAS procs, uses the last group within a categorical variable as the reference group. I included dummy variables for my multilevel categorical predictors, omitting the referent, so I'm not sure the difference in estimates btw nlmixed and fmm can be explained by fmm using the last group as the referent.

amyip · ‎08-29-2024

A follow-up question -- as mentioned I have 6 age groups and 22 counties, so k-1 dummies were included in my model. In defining mu1 and mu0, I should only include one level for age group and county rather than all dummies, correct? Alternatively, if I omit them all, my estimates would be based on these variables taking on the referent value? To simplify, could I define mu1 and m0 as follows: mu1=exp(intercept + b_long_term_care); mu0=exp(intercept); Where the definition of mu includes all the covariates, would mu1 and mu0 assign them all a value of 0 except for mu1 assigning LTC as 1?

amyip · ‎08-28-2024

Thank you so much!

amyip · ‎08-28-2024

You can disregard my earlier post. I corrected my definitions of mu1 and mu0 to depend only on parameters with no variables as follows: mu1=exp(intercept + b_long_term_care*1 + b_female + b_dementia); mu0=exp(intercept + b_female + b_dementia); which yielded the same results as the original model with the additional estimates of mean1= 9.7090, mean0=7.7925, and mean1/mean0=1.2459 with corresponding p-value, lower/upper limits, etc. Can this ratio now be interpreted that, for subjects with all other covariates equal (do they all have to take the value 1?), those in LTC were prescribed on average 1.25x the days' supply as their community-dwelling counterparts? Thanks again for your guidance.

amyip · ‎08-28-2024

I'm beginning to understand -- thank you for your patience! 1-Do I need to include the PARMS statement to set the starting estimate values at something other than 1? In your example, you set b1 at 0 (for the variable MENT). 2-Is it ok to include the parameters as intercept, b_long_term_care, b_female, b_dementia, etc., rather than numbered b0, b1, b2? I have numerous covariates (22 dummies for county) so naming is less confusing than numbering. 3-In your example, you define mu = exp(b0 + b1*ment + b2*fem); ment and fem are data variables, aren't they? is it just mu0 and mu1 that must be functions of parameters and not variables? in which case I would include just the parameters with the values all set at 1 (or 0), correct? Thanks again for your time.

amyip · ‎08-27-2024

proc nlmixed data=ZTNB; y = CumDays; mu = exp(intercept + b_long_term_care*long_term_care + b_female*female + b_dementia*dementia); ll = lgamma(y+1/k) - lgamma(1/k) - lgamma(y+1) + y*log(k*mu) - (y+(1/k))*log(1+k*mu) - log(1-(1+k*mu)**(-1/k)); model y ~ general(ll); mean = mu/(1-(1+k*mu)**(-1/k)); predict mean out=out; mu1=exp(intercept + b_long_term_care*1 + b_female*female + b_dementia*1); mu0=exp(intercept + b_female*female + b_dementia*1); mean1 = mu1/(1-(1+k*mu1)**(-1/k)); mean0 = mu0/(1-(1+k*mu0)**(-1/k)); estimate 'mean LTC=1 at dementia=1' mean1; estimate 'mean LTC=0 at dementia=1' mean0; estimate 'ratio of mean counts: LTC1/LTC0' mean1/mean0; run; This is the code than I ran (full list of model covariates omitted for simplicity) that yielded ERROR: Estimate statement expressions are not allowed to depend on variables in the input data set. I omitted "where CumDays ne 0" because this already applies to all my observations, and the PARMS statement hence the starting value for all parameters is 1. Can you identify the source of the error? Thanks for your time

amyip · ‎08-26-2024

My model includes mostly binary covariates, with dummy variables for 6 age categories and 22 counties; is it correct to omit the referent category for age (include 5 levels) and county (include 21 levels), or should binary dummies for all levels be included? I ran the SAS code that you provided, substituting in my variable names. For the mu1 model I included the term b_long_term_care*1, whereas for the mu0 model I excluded it (b_long_term_care*0), and in both I assigned a value of 1 for one of my covariates (b_dementia*1). The following lines were used verbatim from your example, except for the text in quotation marks which names my variables: mean1=mu1/(1-(1+k*mu1)**(-1/k)); mean0=mu0/(1-(1+k*mu0)**(-1/k)); estimate 'mean fem=1 at ment=10' mean1; estimate 'mean fem=0 at ment=10' mean0; estimate 'ratio of mean counts: fem1/fem0' mean1/mean0;

amyip · ‎08-26-2024

Thanks again for taking the time to follow up. I tried running the model with the estimate statements and got the message: ERROR: Estimate statement expressions are not allowed to depend on variables in the input data set.

amyip · ‎08-22-2024

I am wondering how to interpret estimates from proc nlmixed or proc fmm for zero-truncated negative binomial regression of total cumulative days' supply of a medication class on long-term care (yes/no) after adjusting for other covariates. Using the code below provided by StatDave (thank you!) in another post, I obtain the estimate of 0.2757 (95% CI 0.2254, 0.3261), p<0.0001. Is it appropriate to report the corresponding e^β of 1.32 (95% CI 1.26, 1.39) as a pseudo relative risk, such that long-term care residents were prescribed 1.32x the total days' supply of medication as their community-dwelling counterparts? if so, is there code to include to have these generated with the output? proc nlmixed data=ZTNB; y=CumDays; mean = exp(intercept + b_long_term_care*long_term_care + b_female*female + b_agegr1*agegr1); ll = lgamma(y+1/k) - lgamma(1/k) - lgamma(y+1) + y*log(k*mean) - (y+(1/k))*log(1+k*mean) - log(1-(1+k*mean)**(-1/k)); model y ~ general(ll); run; proc fmm data=ZTNB; model y = long_term_care female agegr1 / dist=truncnegbin; run;

Online Status	Offline
Date Last Visited	2 weeks ago

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

using array to create dummy variable for year

Re: Zero-truncated Negative Binomial Regression

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

Re: using array to create dummy variable for year

using array to create dummy variable for year

Re: Zero-truncated Negative Binomial Regression

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

Re: intertpreting estimates from zero-truncated negative binomial regr...

intertpreting estimates from zero-truncated negative binomial regressi...