Despite me specifying the "message" option in the IC CREATE statement, SAS fails to print the custom error message to the log. Here's my code (note that the red x is an intentional error so that the integrity constraint is triggered): data exampledata; length ID $8 DATE 4 RECORD_QUALITY $1; input ID $ DATE :yymmdd10. RECORD_QUALITY $; format DATE yymmddd10.; datalines; 11111111 2019-04-25 E 22222222 2021-09-16 E 33333333 2021-02-12 F 44444444 2020-05-02 X 55555555 2019-11-21 P ; proc datasets; modify exampledata; ic create id_notnull = not null(ID) message = 'ID must not be missing'; ic create id_unique = unique(ID) message = 'Table must contain one record per ID'; ic create date_check = check(where=(DATE >= '01jan2019'd and DATE <= '31dec2021'd)) message = 'DATE is outside the expected range of 2019-01-01 to 2021-12-31'; ic create record_quality_check = check(where=(RECORD_QUALITY in ('E','F','P'))) message = 'Valid values for RECORD_QUALITY are either E, F, or P'; quit; And here's what I get in the log: Why is SAS not printing my custom error message?     ... View more

Hi, I am trying to conduct an interrupted time series analysis looking at lab test rates (outcome) and how they changed due to the COVID shelter-in-place order in March 2020.    I am making use of this paper to guide me: Time After Time: Difference-in-Differences and Interrupted Time Series Models in SAS I have a individual level dataset of patients with multiple records per patient, one for each month they are in the cohort. From what I understand, since I am looking at testing rates, I could either model this with the individual data as a binary outcome (ie have an flag set to 1 if the patient in a particular month had a test and 0 if they did not) OR I could aggregate the patient level data so that I have the overall testing rate as a percentage per month, meaning I could then model it as a continuous outcome.   Does anyone know what the difference would be for each approach?   From what the linked paper makes it seem like, I should use the GENMOD Procedure if I want to do it as a binary outcome, and use the MIXED Procedure if I want to do it as an aggregated continuous outcome.  The paper provides an example of how to use PROC MIXED for a continuous outcome, but how would I use PROC GENMOD for a binary one? The paper says:   These proportions can be estimated using PROC GENMOD as odds or probabilities on the log scale (relative risks) or as proportions (absolute risks) for each group at each time point. How would one specify if they want relative risks or absolute risks with PROC GENMOD? How does one interpret each? Thank you. ... View more

Hi, I am trying to conduct an interrupted time series analysis looking at lab test rates (outcome) and how they changed due to the COVID shelter-in-place order in March 2020.    I am making use of this paper to guide me: Time After Time: Difference-in-Differences and Interrupted Time Series Models in SAS I have a individual level dataset of patients with multiple records per patient, one for each month they are in the cohort. From what I understand, since I am looking at testing rates, I could either model this with the individual data as a binary outcome (ie have an flag set to 1 if the patient in a particular month had a test and 0 if they did not) OR I could aggregate the patient level data so that I have the overall testing rate as a percentage per month, meaning I could then model it as a continuous outcome.   Does anyone know what the difference would be for each approach?   From what the linked paper makes it seem like, I should use the GENMOD Procedure if I want to do it as a binary outcome, and use the MIXED Procedure if I want to do it as an aggregated continuous outcome.  The paper provides an example of how to use PROC MIXED for a continuous outcome, but how would I use PROC GENMOD for a binary one? The paper says:   These proportions can be estimated using PROC GENMOD as odds or probabilities on the log scale (relative risks) or as proportions (absolute risks) for each group at each time point. How would one specify if they want relative risks or absolute risks with PROC GENMOD? How does one interpret each? Thank you. ... View more