Title


ivtsps -- Two-stage predictor substitution estimators

Syntax

ivtsps depvar [varlist1] (varlist2 = varlist_iv) [if] [in] [weight] [, link(string) log noirr gmm_options]

varlist1 is the list of exogenous variables.

varlist2 is the list of endogenous variables.

varlist_iv is the list of exogenous variables used with varlist1 as instruments for varlist2.

options Description ---------------------------------------------------------------------------------------------- noirr Do not display exponentiated estimates link(string) Link function for the second stage model (identity | logadd | logmult | logit) log Show the GMM iteration log gmm_options gmm##options

Description

ivtsps implements two-stage predictor substitution (TSPS) estimators with several link functions for the second stage model (identity, log-additive, log-multiplicative, logit). It is implemented using generalized method of moments (GMM) estimation by passing the relevant moment condition to the gmm command.

Options

link(identity | logadd | logmult | logit) specifies the link function for the second stage model. identity means the second stage model is a linear regression (which for a binary outcome estimates a causal risk difference). logadd means the second stage model is a Poisson/log-binomial regression (which for a binary outcome estimates a causal risk ratio). logmult means the second stage model is a gamma regression (which for a binary outcome estimates a causal risk ratio, Dukes and Vansteelandt, 2018). logit means the second stage model is a logistic regression (which for a binary outcome estimates a causal odds ratio).

Please see gmm##options

Examples

Read in binary outcome data; y outcome, x exposure, w covariate, z* instrumental variables (genotypes).

. use https://raw.github.com/remlapmot/ivonesamplemr/main/data/ivbinoutdata, clear

Fit the model with a single instrumental variable.

. ivtsps y (x = z1)

Fit the model with multiple instruments.

. ivtsps y (x = z1 z2 z3)

Fit the model with multiple instruments, and adjusting for w.

. ivtsps y w (x = z1 z2 z3)

Using the log additive link function.

. ivtsps y (x = z1 z2 z3), link(logadd)

Using the log multiplicative link function.

. ivtsps y (x = z1 z2 z3), link(logmult)

Using the logit link function.

. ivtsps y (x = z1 z2 z3), link(logit)

Bootstrap standard errors. . bootstrap, reps(250): ivtsps y (x = z1 z2 z3), estonly

Stored results

Please see gmm##results

References

Dukes O, Vansteelandt S. A note on G-estimation of causal risk ratios. American Journal of Epidemiology, 2018, 187, 5, 1079-1084. DOI

Palmer TM, Sterne JAC, Harbord RM, Lawlor DA, Sheehan NA, Meng S, Granell R, Davey Smith G, Didelez V. Instrumental variable estimation of causal risk ratios and causal odds ratios in Mendelian randomization analyses. American Journal of Epidemiology, 2011, 173, 12, 1392-1403. DOI

Terza JV, Basu A, Rathouz PJ. 2008. Two-stage residual inclusion estimation: Addressing endogeneity in health econometric modeling. Journal of Health Economics 27: 531–543. DOI

Author

INCLUDE help ivonesamplemr-author