Measurement error when a variable is in buckets

[Camilla Nytun]

Hello,

I've got a variable, AGE, which in my data set is given in buckets. Meaning someone with AGE=1 is 16-24 and someone with AGE=2 is 25-34 and so on. For the purpose of my study I had to make an assumption about people's ages as assign them a specific age rather than within a bucket. I did this is two ways. Either I made everyone the oldest age in the bucket, meaning everyone with AGE=1 was now 24 OR I also randomly assigned people an age within the bucket.

The data also provides the age at which the individual finished full time studies, AFS. From this I create an estimated work experience variable... AGE-AFS=EXP.

My regression results are significant for both, but very different results. I assume that I need to include this measurement error in my regression, however not sure how to go about doing this.

Thank you

[Nick Cox]

I like the spirit of this, but if I were given data in such form, I would just mutter mildly and use indicator variables for the age categories as delivered.

An extra reason for this is that I wouldn't expect anything of interest to be linear in conventionally measured age in any case.

PS AGE == 6.

[Camilla Nytun]

I like the spirit of this, but if I were given data in such form, I would just mutter mildly and use indicator variables for the age categories as delivered.

An extra reason for this is that I wouldn't expect anything of interest to be linear in conventionally measured age in any case.

PS AGE == 6.

The importance of transforming age into the conventional measure lies in generating a work experience variable by subtracting age and years of full time education.

[Nick Cox]

I appreciate that you want that, but if it could be off by 8 years or more, it is hard to see that any variants improve reliably on taking the data as they arrive.