Table 7

Estimated coefficients of the ordinary least squares (OLS), instrumental variable (IV) and fixed effects (FE) regressions of relative income price (RIP) of cigarettes with interactions of trend and brand types in Bangladesh, 2009–2015

CoefficientRobust SEP valuesCoefficientRobust SEP valuesCoefficientSEP values
Interactions of brands and trend
   Low-price brands
     Wave 20.740.560.180.930.570.113.281.370.02
     Wave 3−0.020.370.960.630.430.150.890.960.35
     Wave 4−2.230.500.00−2.620.850.00−
   Medium-price brands
     Wave 2−1.810.670.01−1.920.670.00−4.361.430.00
     Wave 3−2.970.680.00−−
     Wave 4−5.330.530.00−6.360.650.00−6.321.130.00
   High-price brands
     Wave 2−−−7.661.530.00
     Wave 3−6.441.170.00−9.021.490.00−7.651.220.00
     Wave 4−14.811.310.00−18.502.340.00−14.771.360.00
   Premium brands
     Wave 2−1.632.420.50−1.702.570.51−6.801.680.00
     Wave 3−5.592.740.04−7.943.250.02−12.741.440.00
     Wave 4−22.182.380.00−27.603.890.00−22.321.580.00
 Number of observations588258825882
  • The regression analysis controls for age, gender, education, occupation, socioeconomic status, residence, type of cigarette brands and village fixed effects. The estimates for these control variables are suppressed for the brevity of presentation.

  • Reference categories include male, persons with 0 year of education, owner/tenant farmers, low socioeconomic status, urban area of residence, low cigarette brands and Wave 1 (2009).

  • The SEs of OLS estimates are adjusted for autocorrelation of error terms of multiple observations on the same individual.

  • The village variable is used as an instrument for tax in the IV estimation. The test statistics for exogeneity of the tax variable using the Durbin-Wu-Hausman test is statistically insignificant (robust regression F(1, 2552)=2.04, p=0.1537) suggesting exogeneity. The adjusted R2 from the first-stage regression of tax is 0.9261, suggesting strong predictability of variations in tax using village-level variation.

  • The age variable drops out of the FE regression due to perfect collinearity with the wave variables. Gender, socioeconomic status and area of residence variables are time invariant and hence drop out of the FE regression as well.