"Factor variables may not include negative values" for categorical variable

[Leo Davis]

I am very new to stata and not really sure what I'm doing, I'm just trying to get some regression output for a university assignment.

The data I am using can be found here https://www.datafirst.uct.ac.za/data...hp/catalog/570
"Get microdata" will prompt the download.

The variables I would like to use are named: "w4_a_em1pay" as the dependent variable (monthly income) and "w4_a_em1prod_c" as the independent (economic sector).

w4_a_em1prod_c is a categorical variable with no numerical obs. as far as I know. There are 26000+ observations and I can't check all of them.

I am using the following input:

regress w4_a_em1pay i.w4_a_em1prod_c

And getting the following output:
w4_a_em1prod_c: factor variables may not contain negative values

Is there a fix for this? Can I search for negative values in the variable to remove them? Or somehow ignore the observations with negative values etc.

[Leo Davis]

. table w4_a_em1prod_c

-------------------------------------------------------------------------------------
| Frequency
-------------------------------------------------------------------------+-----------
eb8 - Sector code for primary occupation |
Missing | 71
Private households | 570
Agriculture, hunting, forestry and fishing | 639
Mining and Quarrying | 213
Manufacturing | 685
Electricity, gas and water supply | 66
Construction | 423
Wholesale and Retail trade; repair etc; hotels and restaurants | 1,104
Transport, storage and communication | 286
Financial intermediation, insurance, real estate and business services | 593
Community, social and personal services | 1,784
Total | 6,434

[Ken Chui]

Welcome to Statalist.

Instead of the command used in #2, try this:

table w4_a_em1prod_c, nolab

If there is any negative numeric codes they should be showing up. Once you have identified what they are, you may create a copy of that, and reassign another unique value to the category. For example, this code shows how to replace -8 with 5:

gen w4_2 = w4_a_em1prod_c
replace w4_2 = 5 if w4_2 == -8

The reason of creating a copy is to avoid overwriting the original information. That way, if a mistake was made in the recoding, you'll always have the original w4_a_em1prod_c to go back to.

And if you are absolutely sure that the negative values can be excluded, you may drop those cases with:

regress w4_a_em1pay i.w4_a_em1prod_c if w4_a_em1prod_c >= 0 & w4_a_em1prod_c < .

[Leo Davis]

Welcome to Statalist.

Instead of the command used in #2, try this:

table w4_a_em1prod_c, nolab

If there is any negative numeric codes they should be showing up. Once you have identified what they are, you may create a copy of that, and reassign another unique value to the category. For example, this code shows how to replace -8 with 5:

gen w4_2 = w4_a_em1prod_c
replace w4_2 = 5 if w4_2 == -8

The reason of creating a copy is to avoid overwriting the original information. That way, if a mistake was made in the recoding, you'll always have the original w4_a_em1prod_c to go back to.

And if you are absolutely sure that the negative values can be excluded, you may drop those cases with:

regress w4_a_em1pay i.w4_a_em1prod_c if w4_a_em1prod_c >= 0 & w4_a_em1prod_c < .

Hi Ken, thank you for the fast response.
I tried the command
table w4_a_em1prod_c, nolab and got the following output . table w4_a_em1prod_c, nolab option nolab not allowed r(198); The rest is very helpful, just need to figure out how to identify the relative values

[Rich Goldstein]

please see #4 in https://www.statalist.org/forums/help#adviceextras

[Nick Cox]

Try tabulate rather than table

(I guess you’re using Stata 17 and Ken is using or presuming an earlier version.)

[Ken Chui]

Hi Ken, thank you for the fast response.
I tried the command
table w4_a_em1prod_c, nolab and got the following output . table w4_a_em1prod_c, nolab option nolab not allowed r(198); The rest is very helpful, just need to figure out how to identify the relative values

Oh, sorry. As Nick said in #6, it's just me thinking it was "tab" (as tabulate) while copied and pasted your "table" which is a different command than tab. If you replace "table" with "tab" or "tabulate", then you should be able to see the code instead of label.

tab w4_a_em1prod_c, nolab

[Leo Davis]

Thank you, tab worked

[Richard Williams]

I’m a big fan of the user-written -fre-, which gives you both numeric values and value labels simultaneously. Available from SSC.

[Nick Cox]

We're moving pretty slowly here. Once you see the results you should be able to make a decision on the negative categories. It could be as simple as there is just one negative category, which is for missings. I would change that negative value to a Stata missing value such as system missing (.). See

help mvdecode

If after that negative codes remain for categories you can get a variable that is coded by positive integers by pushing w4_a_em1prod_c through egen group().

Here's a dopey example. Suppose badcat arrived with values -2 -1 0 1 2. You could just add 3 and then everything's positive. But a better idea is to go

egen goodcat = group(badcat), label

and then you get the best of both worlds, as value labels will get carried across.