change 0 to non-missing in certain cases

[Jbraun]

I am trying to do a panel-data analysis and get this message: 

ERROR: Not enough observations with non-missing values for model statement Model 1 in cross section ID=XXXX.

Looking at my data I think it is because the outcome variable has the value 1 for retirement but 0 in all other cases and therefore it probably can't figure out when to consider 0 as an outcome = non-retirement. 

Therefore I want to change all values in outcome to missing if it is zero, except if the last year is 0, then it should keep the zero:

Have:

ID Year Outcome

1  2018 0

1  2019 0

1  2020 1

2  2019 0

2  2020 1

3  2019 0

3  2020 0

3  2021 0

Want:

Have:

ID Year Outcome

1  2018 .

1  2019 .

1  2020 1

2  2019 .

2  2020 1

3  2019 .

3  2020 .

3  2021 0

I hope someone can help with this - probably fairly simple - request. 

Thank you for an awesome community!

Kind regards,

Jacob

[andreas_lds · Accepted solution]

Assuming that the data is sorted by ID and year:

data want;
  set have;
  by id;
  if not last.id and Outcome = 0 then do;
    Outcome = .;
  end;
run;

untested

[Jbraun]

Brilliant, thanks!

[PaigeMiller]

@Jbraun 

Changing zeros to missing can change whatever analysis you are doing and make it gibberish. Without knowing further details about what you are doing, I consider this to be a very poor practice.

(And by the way, an analysis that ignored zeros led to a decision that led to the launch of Space Shuttle Challenger, which then exploded due to a failure of the O-rings because; a grievous error if ever there was one).

[Jbraun]

Yes, can make gibberish or correct the data, depending on the context. I know the NASA case. I promise I will not crash any rockets. 

[ballardw]

I'm going to chime into this late. The Topic of "change 0 to non-missing" and the body with "ERROR: Not enough observations with non-missing values for model statement Model 1 in cross section ID=XXXX." means that assigning additional missing values as requested in the Want data set is not going to address the issue. Adding more missing values means that you have even fewer "non-missing".

The topic title does not reflect the request either.

I would be somewhat interested in seeing the model code you submitted because your narrative makes it sound that the model is poorly designed at best and maybe incorrect completely for the data.

---

@Jbraun wrote:

I am trying to do a panel-data analysis and get this message: 

ERROR: Not enough observations with non-missing values for model statement Model 1 in cross section ID=XXXX.

Looking at my data I think it is because the outcome variable has the value 1 for retirement but 0 in all other cases and therefore it probably can't figure out when to consider 0 as an outcome = non-retirement. 

Therefore I want to change all values in outcome to missing if it is zero, except if the last year is 0, then it should keep the zero:

Have:

ID Year Outcome

1  2018 0

1  2019 0

1  2020 1

2  2019 0

2  2020 1

3  2019 0

3  2020 0

3  2021 0

Want:

Have:

ID Year Outcome

1  2018 .

1  2019 .

1  2020 1

2  2019 .

2  2020 1

3  2019 .

3  2020 .

3  2021 0

I hope someone can help with this - probably fairly simple - request. 

Thank you for an awesome community!

Kind regards,

Jacob

---

[Jbraun]

the title was misleading, I wanted to change 0 to missing. It did not solve the issue, but I found out that I needed to change some missing values in some of the independent variables. It puzzled me a bit because the PROC PANEL documentation says that missing values will just be ignored, but it seems that this was not the case.