interpretation of my triple interaction effect

[Thomas Kok]

Hi everyone. I am doing research on the effect of U.S. tariffs on China. Therefore I used their value added shares in electronics goods consumed in the United States and a weighted average tariff rate. Since my timeline spans over ten years (2012-2022), I added a time dummy to account for higher U.S. tariffs imposed after 2018 on China. As a result, I came up with a triple interaction effect with China and time dummy with tariffs stated below and attached an overview of the results. I only need to know if this has been done correctly, how I can interpret the coefficients of the stata output, what they mean and why the coefficients for the China dummy are that large.

xtreg lnVA c.L_Tariff##i.China##post2018 lnRCA lnWage lnExchange lnPopulation ln_FDI, fe vce (robust)
note: 1.China omitted because of collinearity. Fixed-effects (within) regression

Dependent variable: (log) Country value added component

China interaction effect post-2018
	(1)	(2)	(3)
WATR_(t-1)	-6.472** (2.519)	-6.786** (2.583)	-8.607*** (3.077)
Post-2018 * Tariffs_(t-1)	7.047** (3.169)	7.949** (3.351)	7.484** ( 3.624)
CHN x Tariffs_(t-1)	24.475*** (2.519)	24.516*** (2.913)	27.572*** (4.011)
CHN* Tariffs_(t-1)* Post-2018	-27.122*** (3.169)	-27.734*** (3.856)	-28.515*** (4.735)
Log(RCA)		-.006 (.050)	-.017 (.054)
Log(Wage)		-.016 (.029)	.001 (.030)
Log(FDI)			-.018 (.014)
Log(Population)			-.316 (.462)
Log(Exchange)			.047 (.090)
Constant	2.317*** (.032)	2.479*** (.286)	7.633 (7.797)
Country FE	YES	YES	YES
Observations	630	602	541
Countries	62	62	62
R-squared	0.010	0.003	0.009

Notes: (1) For this sample, we used a fixed effects model. (2) Values in brackets indicate robust standard errors of the regression. (3) Values in parentheses are .^*Significant at the 10% level. * p<.10, ^**Significant at the 5% level. ** p<.05 and ^***Significant at the 1% level. *** p<.01. (4) WATR refers to the weighted average tariff rate.

[Carlo Lazzaro]

Thomas:
1) I would be really concerned about your (within?) Rsq: they are too low to believe that your model gives a reliable representation of the data generating process;
2) the best way to understand interactions (I find three-way interaction too complicated to be explained effectively, though) is to use the - allbaselevels- option (see -help estimation options-).