Doing some work with Wage and Business Income, and have some negative Business Income values. How do I deal with them?

[Caroline Margiotta]

Hi all,

I am trying to run a regression that predicts the proportion of household business and wage income brought home by wives, given age, race, region, and the adjusted business and wage incomes and education levels of husbands and wives. My dataset (from the ASEC supplement of the Current Population Survey) is organised by couples, years, and serial numbers.

I have found, in my dataset, that several self-employed professionals who earn business income are making losses, and that's affecting my Proportion Dependent Variable. I know it's not advisable to simply drop the couples who make negative pre-tax income, but I'm not sure what else I can do.

Could you recommend any procedures that would allow me to create a Proportion of Income_Wives variable without allowing negative Business Income values to adversely affect my results?

Thank you so much!!

[Maarten Buis]

The easiest solution is to just allow negative proportions. In cases with negative buisiness income for women the wife brings in a negative percentage of income and the husband a percentage larger than 100%. This exactly describes the situation you are having.

[Phil Bromiley]

You might want to do something different than Maarten suggests if you believe that most people who generate negative income are actually trying to do something different than people who generate positive income. That is, if there is a substantive difference between the two. For example, some US academics might report negative income from self employment to give themselves a way to deduct office and travel expenditures. Alternatively, someone might have a property really held for capital appreciation or personal consumption but used for business to partially offset related costs. Negative spouse self-employed income might be either of these or even a "hobby" business.

Depending on your precise model and objectives, you might want to treat these differently than conventional employment or self employment. This is difficult since you can have conventional self employment that results in negative income as well. Just for robustness, you might test whether it changes your results if you treat negative income as negative, missing, or zero.