Invalid reference value

[kristiepauly]

Hi All, 

I'm hoping you can help with my code below.  I ran this on Friday and everything was working so I'm not sure why all of a sudden I'm getting this error message. 

I formatted my variables in earlier code. The variable I'm having problems with is diabetes_P1 which I formatted to diabetes_baseline where  0=no and 1 = yes.

Proc format;
value BMI_group
0 - <18.5 = Underweight
18.5 - 24.9 = Healthy
25.0 - <30.0 = Overweight
30.0 - 100.0 = Obese
other = missing;
value diabetes_baseline
0 = No
1 = Yes
other = Missing;
value diabetes_final
0 = No
1 = Yes
other = Missing;

I then wrote code to make new variables to calculate the change in Fev1 and Fev1/FVC over time. The problem is in the diabetes_p1 variable in my prc glm

data working4;   ***this works;
	set working3; 
	change_fev1 = fev1pp_Post_p3 - fev1pp_Post_p1;
	change_fev1_FVC = fev1_FVC_post_p3 - Fev1_FVC_post_p1;
	run;

proc glm data=working4;
	class diabetes_P1(ref= '0') finalgold_P1 (ref='0');
	model change_fev1_FVC = diabetes_P1 finalgold_p1 diabetes_P1*finalgold_p1;
	lsmeans diabetes_P1 finalgold_p1 diabetes_P1*finalgold_p1 / pdiff;
	format finalgold_p1 final_gold_stage. diabetes_P1 diabetes_baseline.;
	run;
ERROR: Invalid reference value for Diabetes_P1.

I tried changing the ref="No" since that is what I formatted it to, but that resulted in the same the error.  

I ran proc contents data=working4;
run;

Alphabetic List of Variables and Attributes
#	Variable	Type	Len	Format	Informat	Label
17	ATS_PackYears_P1	Num	8	 	 	Pack years, from Resp Questionnaire
32	ATS_PackYears_P3	Num	8	 	 	Pack years, from Resp Questionnaire
5	Age_P1	Num	8	 	 	                Age at current visit
19	Age_P3	Num	8	 	 	                Age at current visit
7	BMI_P1	Num	8	 	 	                BMI
21	BMI_P3	Num	8	 	BEST16.	 
34	Change_P1_P3_Smoking_Status	Char	4	 	 Change P1-P3: Smoking Status
12	CortsterOral_P1	Num	8	11.	11.	         Oral corticosteroids
26	CortsterOral_P3	Num	8	 	BEST16.	 
11	Cortsterinhal_P1	Num	8	11.	11.	Inhaled corticosteroids
25	Cortsterinhal_P3	Num	8	BEST12.	BEST16.	 
28	DLco_GLI_pp_PbHb_adj_P3	Num	8	 	 	DLco % predicted GLI, adjusted for Hb and altitude
8	Diabetes_P1	Num	8	11.	11.	         Diabetes

I tried changing the diabetes_P1 variable to diabetes given that is the label attached to it but then get this:

 1          OPTIONS NONOTES NOSTIMER NOSOURCE NOSYNTAXCHECK;
 72         
 73         proc glm data=working4;
 74         class diabetes(ref= 'No') finalgold_P1 (ref='0');
 ERROR: Variable DIABETES not found.
 NOTE: The previous statement has been deleted.
 75         model change_fev1_FVC = diabetes_P1 finalgold_p1 diabetes_P1*finalgold_p1;
 76         lsmeans diabetes_P1 finalgold_p1 diabetes_P1*finalgold_p1 / pdiff;
 ERROR: Only CLASS variables allowed in this effect.
 NOTE: The previous statement has been deleted.
 77         format finalgold_p1 final_gold_stage. diabetes_P1 diabetes_baseline.;
 78         run;

Any suggestions? 

[FreelanceReinh]

Hi @kristiepauly,

Reference values must always be specified as formatted values, so class diabetes_P1(ref='No') ... should work.

[kristiepauly]

Hi @FreelanceReinh. I tried that.  It did not work. 

[FreelanceReinh · Accepted solution]

Then please provide more details of that failed attempt, e.g., the complete log of the PROC GLM step.

What you've shown so far is

•  diabetes_P1(ref= '0') -- wrong reference value because '0' is not among the labels (=formatted values) of format diabetes_baseline. (i.e. 'No', 'Yes').
• diabetes(ref= 'No') -- wrong variable name, you cannot use the variable label here.

If diabetes_P1(ref='No') is rejected with the same error message "Invalid reference value ...", then your PROC FORMAT step might contain invisible characters. Note that format labels should normally be quoted:

proc format;
value BMI_group
0 - <18.5   = "Underweight"
18.5 - 24.9 = "Healthy"
...

although omitting the quotes works in most cases. Without the quotes you cannot see the difference, e.g., between "No" and "No ", where the trailing blank in the latter is in fact a non-breaking space character. In this case ref='No' would not work because of that invisible third character in the formatted value.

[kristiepauly]

@FreelanceReinh @Quentin . I deleted the code and rewrote and it works fine now.  Gotta love SAS /s.  Thank you! 

[Quentin]

Interesting, ref='No' Should work.  Maybe double-check the capitalization in your format and ref= to make sure they match?  This trivial example works for me:

proc format ;
  value age
    low-13='No'
    14-high='Yes'
   ;
run ;

proc glm data=sashelp.class ;
  class age(ref='No') ;
  model height=age ;
  format age age. ;
run ;
quit ;

Can you post the full log from your PROC FORMAT step and PROC GLM step with ref='No'?

[Patrick]

When things stop working or for a final test make it a habit to create a new SAS session and re-run everything. This helps you avoid that something you've changed in a previous run in the same SAS session impacts your result.

Just on a side note: Your format BMI_group will return MISSING for values like 24.95 I assume that's not what you want.

proc format;
  value BMI_group
    0 - <18.5 = Underweight
    18.5 - 24.9 = Healthy
    25.0 - <30.0 = Overweight
    30.0 - 100.0 = Obese
    other = missing;

  value BMI_group_new
    0    -< 18.5 = 'Underweight'
    18.5 -< 25   = 'Healthy'
    25   -< 30   = 'Overweight'
    30   - 100   = 'Obese'
    other        = 'missing'
    ;
run;

data test;
  input val;
  BMI_group     =put(val,bmi_group.);
  BMI_group_new =put(val,bmi_group_new.);
  datalines;
18.49
18.5
24.89
24.9
24.99
25
29.99
30
30.01
;
run;

proc print data=test;
run;

[kristiepauly]

@Patrick Thank you for picking up on that.  🙂