Watch this Ask the Expert session to learn how to use SAS survey tools to design and analyze big survey data scientifically.
Watch the webinar
Join George Fernandez as he discusses why survey analytics differs from other standard statistical procedures. He will cover data exploration, description, model building and interpretation. Watch this webinar to learn:
- Why standard statistical procedures are inappropriate for analyzing probability surveys.
- How to use advanced features included in the SAS survey procedures.
- How to explore, summarize, fit the model and interpret results using SAS survey procs.
- How to use the popular NHANES survey data in a live demo of the advanced features of SAS survey procs.
The questions from the Q&A segment held at the end of the webinar are listed below and the slides from the webinar are attached.
Do you know if SAS plans to add a new survey procedure to fit Poisson regression?
I don’t think SAS is going to add a new Survey procedure to handle Poisson regression directly, but please follow the directions provided in these links:
https://support.sas.com/rnd/app/stat/examples/SurveyPoisson/surveypoisson.htm
https://support.sas.com/rnd/app/stat/examples/SurveyPoisson/sas.html
If I want to do analysis on a sub-population, such as ages 18-64 and employed at time of interview, is it okay to use a where statement or is the only correct option to use domain?
You should always use domain statement for sub-population analysis unless you modified the target population.
Why does PROC SURVEYMEANS not report Standard deviations?
PROC SURVEYMEANS computes the NOBS, MEAN, STDERR, and CLM statistics by default, but you can request standard deviation by adding STD as an option. Please check the documentation for additional keywords.
Does SAS have the option to specify the type of adjustment (ie TUKEY) when comparing the domain means?
Unfortunately, Only Bonferroni adjustment is available in SURVEYMEANS Domain statement for mean comparison.
Can PROC SURVEYFREQ do a Cochran-Armitage trend test?
Unfortunately, Cochran-Armitage trend test is not included with PROC SURVEYFREQ at this time.
How do you compare the domain means (get a p.val) when using ratio estimation of PROC SURVEYMEANS? E.g., If we know one ratio variable's mean and sex in women and men (sex as a domain), how can we get the p.val between them?
Comparing Ratio means is implemented using the SMSUB SAS macro. Please contact SAS Technical Support for more information about using this SAS macro using SAS IML.
How would you approach testing for between group differences for categorical variables and a 3 x 3 table?
You could try to fit a Generalized logit model using PROC SURVEYLOGISTIC after collapsing 3 x 3 two-way combinations to one-way with 9 levels and using LSMEANS or LSMESTIMATE statements in PROC SURVEYLOGISTIC.
Could PROC MIXED be used to analyze survey data with design and stratification weights?
PROC MIXED is not supported for Survey analysis. But PROC GLIMMIX is supported. Please watch https://www.youtube.com/watch?v=VE_HM8JcMZM
And refer to this document: https://support.sas.com/resources/papers/proceedings14/SAS026-2014.pdf
Will survey procedure be implemented specifically for the propensity score matching?
Not within Survey procs. Please refer the new SAS STAT procedure PSMATCH documentation at https://support.sas.com/documentation/onlinedoc/stat/142/psmatch.pdf for propensity score matching in general.
Can you please explain more under condition(s) to use the different variance estimation methods, e.g. Taylor series, jackknife, etc.
Tyler series method is the default method in all survey procs. You need to always include survey design details with Tyler method. If you use the replication-based methods (BRR, JK etc.), after generating replicate weights, you can omit the survey design (Cluster, strata) details. This is the most significant advantage with these replication-based weights. But the computation time may be significantly higher with big data.
Do we need SUDAAN if we have the SAS SURVEY Procedures? Can the SURVEY procedures do everything that SUDAAN can do?
SAS SURVEY procedures can do almost everything SUDAAN can do. If you want to check on something, please just search sas.com to check for documentation on the item in question.
Can SAS calculate temporal trend for survey data?
Not directly within the SAS Survey procs. You can try Proc GLIMMIX with time correlated covariance structure.
Can you not use where/by statements with the BRR variance method?
Where or By statement is not recommended with Designed surveys. Try to use domain statement for subpopulation analysis with BRR or JK methods.
Will PROC SURVEYIMPUTE create multiple imputed complete data and export for later analyses?
Multiple imputed data sets are only needed when used with PROC MI/MINANALYSE procs. SURVEYIMPUTE generates replication-based weights and these datasets can be saved and used subsequently with any other SAS survey analysis procedures.
Why is it so hard to estimate standard error using survey data?
Because when survey data are collected based on specific survey designs (not just from random sample) and estimation and inferences are needed for finite population we must use specialized methods to estimate variances and covariances. That is why we use special survey procs within SAS/STAT to estimate the standard errors.
When using PROC SURVEYIMPUTE, do replication weights also need to be used?
Yes, you need to use replication weights because SAS is using replication-based variance estimation methods when imputing missing values.
Do you need to set up the number of imputed data that you want to obtain with PROC SURVEYIMPUTE?
Number of imputed data sets is only needed when used with PROC MI and PROC MINANALYSE.
Could you speak about the different options for confidence intervals? My colleagues were recently discussing using type=logit in PROC SURVEYFREQ.
You can specify TYPE=Logit with PROC SURVEYFREQ. This specifies the type of confidence limits to compute for proportions. If you do not specify the TYPE= cl-option, PROC SURVEYFREQ computes Wald confidence limits (TYPE=WALD) by default.
You can specify one of the following confidence limit types:
LOGIT = requests logit confidence limits for proportions. If you specify the CL(TYPE=LOGIT) option, PROC SURVEYFREQ computes logit confidence limits for proportions. For more information, see Agresti (2013) and Korn and Graubard (1998).
I want to learn SAS. What is the price for it?
You can learn more about SAS on our website. We also have some free training available right now that I encourage you to explore.
Is SAS planning to add more diagnostic such as goodness of fit test in SURVEYLOGISTIC?
Please contact SAS Technical Support for more information about the SURVEYLOGISTIC procedure.
Recommended Resources
Wait Wait, Don't Tell Me… You're Using the Wrong Proc!
SAS/STAT® 14.3 User’s Guide Introduction to Survey Sampling and Analysis Procedures
The Latest and Greatest Capabilities of the SURVEY Procedures
Want more tips? Be sure to subscribe to the Ask the Expert board to receive follow up Q&A, slides and recordings from other SAS Ask the Expert webinars.
Click image to register for webinar