Adding Standard errors on graphs

[Anita_n]

Dear all,

I have a graph showing the mean ages in different diagnosis years. I wish to add standard errors to this plot. Is it right to do that with proc means like below and then deduct or add these values from/to the maen ages to get the upper and lower limits. Or do I say lower= mean_age- 2*stderr ,

and upper= mean_age+2*stderr. Why is this standard error important?

proc means data=mydata noprint;
var age;
class year sex;
output out = myoutdat mean=mean_age stderr=stderr ;
run;

[Ksharp]

proc means data=sashelp.class noprint;
var age;
class sex;
output out = myoutdat mean=mean_age stderr=stderr ;
run;

data myoutdat;
set myoutdat;
lower= mean_age- 2*stderr ;
higher= mean_age+ 2*stderr ;
run;
proc sgplot data=myoutdat;
scatter x=sex y=mean_age/yerrorlower=lower yerrorupper=higher;
run;

[Anita_n]

Thanks, I just wanted to know which solution is better, is it better using the LCLM  and UCLM  in proc means or using mean_age +/- 2*stderr to get the confident intervals

[ballardw]

---

@Anita_n wrote:

Thanks, I just wanted to know which solution is better, is it better using the LCLM  and UCLM  in proc means or using mean_age +/- 2*stderr to get the confident intervals

---

Compare the two values for some data. See which makes sense for your purpose. There's also CLM that you could add/subtract from the mean to get a limit to display.

Sometimes one of the limits may not make sense and needs to be censored. For instance if your data relates to patient weight and you get a LCLM of -5 (pounds or kilograms) does that make sense? A weight for a person in the negative range is highly problematic (and likely would come from data with extreme values, small sample and possibly a mix of adults and children).

So "best" comes from "know your data" and making decisions to show the story you intend.

[Anita_n]

@ballardw yes, you are right I compared the values and realised they are very similar, there just very very small differences. In my case there were no negative values for the mean ages. I think I will just choose one of the methods

[Ksharp]

I would prefer to " LCLM and UCLM ".