Taking the ratio of a Y variable in a linear regression with a dummy X

[Emm Ra]

I am trying to figure out how to divide the values of the variable when my X=1 versus when my X=0 rather than subtracting, as a linear regression does.
For example, I have a Y variable that takes on values between -30 and +30. I have a dummy X variable taking on values of 0 and 1. When I run a simple linear regression, I get the following:
reg Y X
where the coefficient on X is equal to the value of Y when X=1 MINUS the value of Y when X=0.
What I need instead is for the coefficient on X to isolate the value of Y when X=1 DIVIDED BY the value of Y when X=0, and for the standard error to be calculated properly. Unfortunately I cannot simply transform the Y variable to be equal to a ratio because this is a between-subjects design (some subjects have X=0, some have X=1).
My first thought was to transform the Y variable via logs or something else, but the negative values of the variable make logs a suboptimal choice.
I'm looking for other transformations or alternative regressions that can accomplish this! Thanks in advance for any thoughts. It seems like there should be an obvious solution but it's not coming to me.

[Clyde Schechter]

You can try a general linear model with log-link, and then exponentiate the coefficient.

glm Y i.X, link(log)
display "Ratio = " =exp(_b[1.X])

But it may fail to converge, in which case I don't know what else to suggest.

To be honest, it's hard for me to imagine a situation where it is meaningful to compare the ratios of things whose range crosses over 0. Maybe if you explain your context, some other approach might become more apparent.

[Joseph Coveney]

and for the standard error to be calculated properly.

With Clyde's suggestion, you can use the -eform- option of -glm- to get at the standard error. Or you can use -nlcom- afterward. If you have trouble with convergence (a real possibility, given that you have negative values), then you can use -regress- for a noniterative fit, again, followed by -nlcom- to get the ratio version of the coefficient and its standard error. See below.

.ÿ
.ÿversionÿ17.0

.ÿ
.ÿclearÿ*

.ÿ
.ÿsetÿseedÿ`=strreverse("1605135")'

.ÿ
.ÿquietlyÿsetÿobsÿ50

.ÿgenerateÿbyteÿxÿ=ÿ_nÿ>ÿ_Nÿ/ÿ2

.ÿgenerateÿdoubleÿyÿ=ÿrnormal(25ÿ+ÿ25ÿ*ÿx,ÿ0.5)

.ÿ
.ÿversionÿ16.1:ÿtableÿx,ÿcontents(meanÿy)

----------------------
ÿÿÿÿÿÿÿÿxÿ|ÿÿÿÿmean(y)
----------+-----------
ÿÿÿÿÿÿÿÿ0ÿ|ÿÿ25.108971
ÿÿÿÿÿÿÿÿ1ÿ|ÿÿ49.896089
----------------------

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿglmÿyÿi.x,ÿfamily(gaussian)ÿlink(log)ÿeformÿnolog

GeneralizedÿlinearÿmodelsÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿ=ÿÿÿÿÿÿÿÿÿ50
Optimizationÿÿÿÿÿ:ÿMLÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿResidualÿdfÿÿÿÿÿ=ÿÿÿÿÿÿÿÿÿ48
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿScaleÿparameterÿ=ÿÿÿ.2324261
Devianceÿÿÿÿÿÿÿÿÿ=ÿÿ11.15645483ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ(1/df)ÿDevianceÿ=ÿÿÿ.2324261
Pearsonÿÿÿÿÿÿÿÿÿÿ=ÿÿ11.15645483ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ(1/df)ÿPearsonÿÿ=ÿÿÿ.2324261

Varianceÿfunction:ÿV(u)ÿ=ÿ1ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ[Gaussian]
Linkÿfunctionÿÿÿÿ:ÿg(u)ÿ=ÿln(u)ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ[Log]

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿAICÿÿÿÿÿÿÿÿÿÿÿÿÿ=ÿÿÿ1.417872
Logÿlikelihoodÿÿÿ=ÿ-33.44680749ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿBICÿÿÿÿÿÿÿÿÿÿÿÿÿ=ÿÿ-176.6206

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿOIM
ÿÿÿÿÿÿÿÿÿÿÿyÿ|ÿÿÿÿÿexp(b)ÿÿÿstd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿ1.xÿ|ÿÿÿ1.987182ÿÿÿ.0085427ÿÿÿ159.74ÿÿÿ0.000ÿÿÿÿÿ1.970509ÿÿÿÿ2.003996
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ25.10897ÿÿÿ.0964212ÿÿÿ839.36ÿÿÿ0.000ÿÿÿÿÿÿ24.9207ÿÿÿÿ25.29867
------------------------------------------------------------------------------

.ÿnlcomÿRatio:ÿexp(_b[1.x])

ÿÿÿÿÿÿÿRatio:ÿexp(_b[1.x])

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿyÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿRatioÿ|ÿÿÿ1.987182ÿÿÿ.0085427ÿÿÿ232.62ÿÿÿ0.000ÿÿÿÿÿ1.970438ÿÿÿÿ2.003925
------------------------------------------------------------------------------

.ÿ
.ÿ//ÿIfÿyouÿhaveÿtroubleÿwithÿconvergence,ÿthen
.ÿregressÿyÿibn.x,ÿnoconstant

ÿÿÿÿÿÿSourceÿ|ÿÿÿÿÿÿÿSSÿÿÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿMSÿÿÿÿÿÿNumberÿofÿobsÿÿÿ=ÿÿÿÿÿÿÿÿ50
-------------+----------------------------------ÿÿÿF(2,ÿ48)ÿÿÿÿÿÿÿÿ>ÿÿ99999.00
ÿÿÿÿÿÿÿModelÿ|ÿÿ78002.0032ÿÿÿÿÿÿÿÿÿ2ÿÿ39001.0016ÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿ=ÿÿÿÿ0.0000
ÿÿÿÿResidualÿ|ÿÿ11.1564548ÿÿÿÿÿÿÿÿ48ÿÿ.232426142ÿÿÿR-squaredÿÿÿÿÿÿÿ=ÿÿÿÿ0.9999
-------------+----------------------------------ÿÿÿAdjÿR-squaredÿÿÿ=ÿÿÿÿ0.9999
ÿÿÿÿÿÿÿTotalÿ|ÿÿ78013.1597ÿÿÿÿÿÿÿÿ50ÿÿ1560.26319ÿÿÿRootÿMSEÿÿÿÿÿÿÿÿ=ÿÿÿÿ.48211

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿyÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿxÿ|
ÿÿÿÿÿÿÿÿÿÿ0ÿÿ|ÿÿÿ25.10897ÿÿÿ.0964212ÿÿÿ260.41ÿÿÿ0.000ÿÿÿÿÿÿ24.9151ÿÿÿÿ25.30284
ÿÿÿÿÿÿÿÿÿÿ1ÿÿ|ÿÿÿ49.89609ÿÿÿ.0964212ÿÿÿ517.48ÿÿÿ0.000ÿÿÿÿÿ49.70222ÿÿÿÿ50.08996
------------------------------------------------------------------------------

.ÿnlcomÿRatio:ÿ_b[1.x]ÿ/ÿ_b[0.x]

ÿÿÿÿÿÿÿRatio:ÿ_b[1.x]ÿ/ÿ_b[0.x]

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿyÿ|ÿCoefficientÿÿStd.ÿerr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿconf.ÿinterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿRatioÿ|ÿÿÿ1.987182ÿÿÿ.0085427ÿÿÿ232.62ÿÿÿ0.000ÿÿÿÿÿ1.970438ÿÿÿÿ2.003925
------------------------------------------------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

By the way, I wholeheartedly agree with Clyde. Are your outcome variable's values always one sign when the value of the predictor is zero and always the other sign when the predictor is one? Even then, I'm not sure I'd know what to make of the ratio.