PROC MI MNAR adjustobs

[_Hopper]

When using PROC MI I was provided some code but it does not work. In my dataset I have three treatment values (1 2 3).

There are two MNAR calls in the code, one is for TRT  = 3 and the other is for TRT 1 and 2 combined. PROC MI says the following code is not allowed because a symbol other an equal "=" isn't allowed. Any thoughts on how to make this work are appreciated.

proc mi data=&data seed=123 nimpute=100 out=outmi;
class trtpn;
fcs reg;
mnar adjust (m12/shift=&k1 adjustobs=(trtpn^=3));<<<<<<<<<<<<<<<<<<<<<<-this line causes an error
mnar adjust (m12/shift=&k2 adjustobs=(trtpn=3));
var m3 m6 m12 m24 m36 ;
run;

[sbxkoenk]

I think @SAS_Rob can help you out.

(he's being notified due to me @-mentioning him here)

[SAS_Rob]

MI will only allow you to use an = sign in adjustobs= option. 

My suggestion would be to create a new variable that only has two levels to it, the combined 1,2 and 3 using a data step by defining it something like this:

trtpn_new=(trtpn ne 3);

Then use that variable in the Proc MI step.

proc mi data=&data seed=123 nimpute=100 out=outmi;
class trtpn_new;
fcs reg;
mnar adjust (m12/shift=&k1 adjustobs=(trtpn_new=1));
mnar adjust (m12/shift=&k2 adjustobs=(trtpn_new=0));
var m3 m6 m12 m24 m36 ;
run;

[_Hopper]

Rob,

Thanks for the suggestion. Would something like this work as an alternative:

mnar adjust (m12/shift=&k1 adjustobs=(treatnp='1' '2'));
mnar adjust (m12/shift=&k2 adjustobs=(treatnp='3'));

I'm fine using your solution as well.

[SAS_Rob · Accepted solution]

Yes, alternatively, you can use a list inside the parenthesis.  This approach in your case might be easier so I would recommend that.

mnar adjust (m12/shift=&k1 adjustobs=(treatnp='1' '2'));

[_Hopper]

Rob - thanks. One other item isn't quite working for me.

Let's say my change mean for treatment groups= -100

Let's say my control change mean = -200

I set S1 = -100 and S2 = -200 then my increment is +15 if I want 20 equal intervals. Is this correct?

The MI output doesn't look right - the shift values appear to be going the wrong way as the model iterates. Any thoughts? 

[SAS_Rob]

Can you post the most recent code along with the output that shows what you are seeing?

[_Hopper]

Rob,

As it turns out the model code I was given contained an error, so that S2 was not properly computed. Once I corrected it, the code ran as expected.

Thanks for your help!