Skip to main content

A propensity score adjustment method for longitudinal time series models under nonignorable nonresponse


Analysis of data with nonignorable nonresponse is an important and challenging task. Although some methods have been developed for inference under nonignorable nonresponse, they are only available for independent data. In this paper, we develop a two-stage propensity score adjustment method to estimate longitudinal time series models with nonignorable missingness. In particular, the response probability or propensity score is first estimated via solving the mean score equation based on the observed sample. Then, the inverse propensity scores are employed to conduct weighting adjustment for a composite likelihood based estimation. The propensity scores weighted estimation equations are shown to yield consistent and asymptotic normal estimators. Simulation studies and application to AIDS Clinical Trial data are presented to evaluate the performance of the proposed method.

This is a preview of subscription content, access via your institution.


  1. Bahari F, Parsi S, Ganjali M (2019) Empirical likelihood inference in general linear model with missing values in response and covariates by MNAR mechanism. Stat Pap.

    Article  Google Scholar 

  2. Bickel PJ, Doksum KA (1977) Mathematical statistics: basic ideas and selected topics. Holden-Day, San Francisco

    MATH  Google Scholar 

  3. Binder DA (1983) On the variances of asymptotically normal estimators from complex surveys. Int Stat Rev 51(3):279–292

    MathSciNet  Article  Google Scholar 

  4. Brockwell PJ, Davis RA (1991) Time series: theory and methods, 2nd edn. Springer, New York

    Book  Google Scholar 

  5. Davis RA, Yau CY (2011) Comments on pairwise likelihood in time series models. Stat Sin 21(1):255–277

    MathSciNet  MATH  Google Scholar 

  6. Da Silva DN, Opsomer JD (2009) Nonparametric propensity weighting for survey nonresponse through local polynomial regression. Surv Methodol 35(2):165–176

    Google Scholar 

  7. Da Silva DN, Opsomer JD (2006) A kernel smoothing method of adjusting for unit non-response in sample surveys. Can J Stat 34(4):563–579

    MathSciNet  Article  Google Scholar 

  8. Eideh AAH, Nathan G (2006) Fitting time series models for longitudinal survey data under informative sampling. J Stat Plan Inference 136(9):3052–3069

    MathSciNet  Article  Google Scholar 

  9. Folsom, R. E. (1991). Exponential and logistic weight adjustments for sampling and nonresponse error reduction. Proceedings of the Social Statistics Section, American Statistical Association, pp 197–202

  10. Härdle W (1990) Applied nonparametric regression. Cambridge University Press, Boston

    Book  Google Scholar 

  11. Jiang DP, Zhao PY, Tang NS (2016) A propensity score adjustment method for regression models with nonignorable missing covariates. Comput Stat Data Anal 94:98–119

    MathSciNet  Article  Google Scholar 

  12. Joe H, Lee Y (2009) On weighting of bivariate margins in pairwise likelihood. J Multivar Anal 100(4):670–685

    MathSciNet  Article  Google Scholar 

  13. Kim JK, Im J (2014) Propensity score adjustment with several follow-ups. Biometrika 101(2):439–448

    MathSciNet  Article  Google Scholar 

  14. Kim JK, Riddles MK (2012) Some theory for propensity-score-adjustment estimators in survey sampling. Surv Methodol 38(2):157–165

    Google Scholar 

  15. Kim JK (2011) Parametric fractional imputation for missing data analysis. Biometrika 98:119–132

    MathSciNet  Article  Google Scholar 

  16. Kim JK, Kim JJ (2007) Nonresponse weighting adjustment using estimated response probability. Can J Stat 35(4):501–514

    MathSciNet  Article  Google Scholar 

  17. Little RJA, Rubin DB (2002) Statistical analysis with missing data. Wiley, New York

    Book  Google Scholar 

  18. Little RJA (1986) Survey nonresponse adjustments for estimates of means. Int Stat Rev 54(2):139–157

    Article  Google Scholar 

  19. Little RJA (1988) Missing-data adjustments in large surveys. J Bus Econ Stat 6(3):287–296

    Google Scholar 

  20. Liu Q, Pierce DA (1994) A note on Gauss-Hermite quadrature. Biometrika 81(3):624–629

    MathSciNet  MATH  Google Scholar 

  21. Liu T, Yuan X (2018) Doubly robust augmented-estimating-equations estimation with nonignorable nonresponse data. Stat Pap.

    Article  MATH  Google Scholar 

  22. Ng CT, Joe H, Karlis D, Liu J (2011) Composite likelihood for time series models with a latent autoregressive process. Stat Sin 21(1):279–305

    MathSciNet  MATH  Google Scholar 

  23. Priya RD, Kuppuswami S, Sivaraj R (2015) Bayesian based inference of missing time series values using Genetic Algorithm. Int J Hybrid Intell Syst 12:77–87

    Google Scholar 

  24. Qin J, Leung D, Shao J (2002) Estimation with survey data under nonignorable nonresponse or informative sampling. J Am Stat Assoc 97(457):193–200

    MathSciNet  Article  Google Scholar 

  25. Rosenbaum PR, Rubin DB (1983) The central role of the propensity score in observational studies for causal effects. Biometrika 70(1):41–55

    MathSciNet  Article  Google Scholar 

  26. Riddles MK, Kim JK, Im J (2016) A propensity-score-adjustment method for nonignorable nonresponse. J Surv Stat Methodol 4(2):215–245

    Article  Google Scholar 

  27. Rubin DB (1976) Inference and missing data. Biometrika 63(3):581–592

    MathSciNet  Article  Google Scholar 

  28. Stubbendick AL, Ibrahim JG (2006) Likelihood-based inference with nonignorable missing responses and covariates in models for discrete longitudinal data. Stat Sin 16:1143–1167

    MathSciNet  MATH  Google Scholar 

  29. Shao J, Wang L (2016) Semiparametric inverse propensity weighting for nonignorable missing data. Biometrika 103(1):175–187

    MathSciNet  Article  Google Scholar 

  30. Silverman BW (1986) Density estimation for statistics and data analysis. Chapman and Hall, London

    Book  Google Scholar 

  31. Tang NS, Zhao PY, Zhu HT (2014) Empirical likelihood for estimating equations with nonignorably missing data. Stat Sin 24:723–747

    MathSciNet  MATH  Google Scholar 

  32. Tseng CH, Elashoff R, Li N, Li G (2016) Longitudinal data analysis with non-ignorable missing data. Stat Methods Med Res 25(1):205–220

    MathSciNet  Article  Google Scholar 

  33. Van Der Vaart AW (1998) Asymptotic statistics. Cambridge University Press, Cambridge

    Book  Google Scholar 

  34. Vasdekis VGS, Rizopoulos D, Moustaki I (2014) Weighted pairwise likelihood estimation for a general class of random effects models. Biostatistics 15(4):677–689

    Article  Google Scholar 

  35. Wang L, Qi CC, Shao J (2019) Model-assisted regression estimators for longitudinal data with nonignorable dropout. Int Stat Rev 87(S1):S121–S138

    MathSciNet  Article  Google Scholar 

  36. Yuan Y, Yin GS (2010) Bayesian quantile regression for longitudinal studies with nonignorable missing data. Biometrics 66:105–114

    MathSciNet  Article  Google Scholar 

  37. Zhang H, Paik MC (2009) Handling missing responses in generalized linear mixed model without specifying missing mechanism. J Biopharm Stat 19(6):1001–1017

    MathSciNet  Article  Google Scholar 

  38. Zhang GY, Yuan Y (2012) Bayesian modelling longitudinal dyadic data with nonignorable dropout, with application to a breast cancer study. Ann Appl Stat 6(2):753–771

    MathSciNet  Article  Google Scholar 

  39. Zhang W, Xie F, Tan J (2020) A robust joint modeling approach for longitudinal data with informative dropouts. Comput Stat.

    MathSciNet  Article  MATH  Google Scholar 

  40. Zhao PY, Wang L, Shao J (2018) Analysis of longitudinal data under nonignorable nonmonotone nonresponse. Stat Interface 11(2):265–279

    MathSciNet  Article  Google Scholar 

  41. Zhou M, Kim JK (2012) An efficient method of estimation for longitudinal surveys with monotone missing data. Biometrika 99:631–648

    MathSciNet  Article  Google Scholar 

Download references


The authors thank the Editor, the Associate Editor and referees for their constructive comments. The collaborative work described in this paper was supported by HKSAR-RGC-GRF Nos 14305517, 14601015 and 14302719 (Yau) and National Social Science Foundation of China, No. 18BTJ022 (Liu).

Author information



Corresponding author

Correspondence to Zhan Liu.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.



Proof of the response model identifiability. For the response model identifiability, the mean score function \(\bar{S}(\eta )\) is not identifiable if there exist different \(\eta \) and \(\eta '\) such that \(\bar{S}(\eta )=\bar{S}(\eta ')\) for all \(x_{i}\), \(\delta _{i,t}\) and \(y_{i,t}\). To be specific, the mean score function \(\bar{S}(\eta )\) in (3.7) consists of two parts, \(\sum \nolimits _{i=1}^{n}\sum \nolimits _{t=2}^{T}\delta _{i,t}s_{y_{i,t}}(\eta )\) and

$$\begin{aligned} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}(1-\delta _{i,t})\frac{\int \pi ^{-1}_{y}(\eta )\hat{f}(y|x_{i},\delta _{i,t}=1)\partial \pi _{y}(\eta )/\partial \eta dy}{\int (\pi ^{-1}_{y}(\eta )-1)\hat{f}(y|x_{i},\delta _{i,t}=1)dy} \,. \end{aligned}$$

The first part is not identifiable if and only if two different \(\eta \) and \(\eta '\) give the same \(s_{y_{i,t}}(\eta )=\{\delta _{i,t}-\pi _{i,t}(\eta )\}\pi ^{-1}_{i,t}(\eta )\{1-\pi _{i,t}(\eta )\}^{-1}\partial \pi _{i,t} (\eta )/\partial \eta \) for all possible values of \(x_{i}\), \(\delta _{i,t}\) and \(y_{i,t}\), where \(\pi _{i,t} (\eta )\) is defined by (2.2). Here, we take the derivative of \(\pi _{i,t}(\eta )\) with respect to \(\eta _{1}\) for example, then we have

$$\begin{aligned} s_{y_{i,t}}(\eta )&=\{\delta _{i,t}-\pi _{i,t}(\eta )\}\pi ^{-1}_{i,t}(\eta )\{1-\pi _{i,t}(\eta )\}^{-1}\partial \pi _{i,t} (\eta )/\partial \eta _{1} \nonumber \\&= \frac{x_{i}[\delta _{i,t}\exp (-\eta _{1}x_{i}-\eta _{2}y_{i,t})+\delta _{i,t}-1]}{1+\exp (-\eta _{1}x_{i}-\eta _{2}y_{i,t})} \,. \end{aligned}$$

Suppose that there exist that \(\eta =(\eta _{1}, \eta _{2})\) and \(\eta '=(\eta '_{1}, \eta '_{2})\) such that

$$\begin{aligned} \frac{x_{i}[\delta _{i,t}\exp (-\eta _{1}x_{i}-\eta _{2}y_{i,t})+\delta _{i,t}-1]}{1+\exp (-\eta _{1}x_{i}-\eta _{2}y_{i,t})}&=\frac{x_{i}[\delta _{i,t}\exp (-\eta '_{1}x_{i}-\eta '_{2}y_{i,t})+\delta _{i,t}-1]}{1+\exp (-\eta '_{1}x_{i}-\eta '_{2}y_{i,t})} \end{aligned}$$

for all values of \(x_{i}\), \(\delta _{i,t}\) and \(y_{i,t}\). It is easy to see that \(\eta '_{1}=\eta _{1}\) must hold. Also, if \(\eta '_{1}=\eta _{1} \ne 0\), then by taking \(x_{i}\) sufficiently large or negative, we have \(\delta _{i,t}\exp (-\eta _{1}x_{i}-\eta _{2}y_{i,t})+\delta _{i,t}-1=\delta _{i,t}\exp (-\eta '_{1}x_{i}-\eta '_{2}y_{i,t})+\delta _{i,t}-1\). It follows that \(\eta '_{2}=\eta _{2}\). That is, the first part is identifiable.

The second part in (A1) is not identifiable if and only if two different \(\eta \) and \(\eta '\) give the same \(\pi ^{-1}_{y}(\eta )\partial \pi _{y}(\eta )/\partial \eta = \pi ^{-1}_{y}(\eta ')\partial \pi _{y}(\eta ')/\partial \eta \) and \(\pi ^{-1}_{y}(\eta )-1=\pi ^{-1}_{y}(\eta ')-1\) for all possible values of \(x_{i}\), \(\delta _{i,t}\) and \(y_{i,t}\), because \(\hat{f}(y|x_{i},\delta _{i,t}=1)\) in (A1) is obtained by (3.6) and not related to the parameter \(\eta \). Similar to the discussion in the first part, we have that two different \(\eta \) and \(\eta '\) cannot produce the equal second part. That is, the second part is also identifiable. Thus, the response model parameter \(\eta \) estimated by the mean score equation in (3.7) is identifiable.

Proof of Lemma 1

Note that

$$\begin{aligned} \mathcal {I}_{21}(\theta , \eta )=&-E\bigg \{ \frac{\partial u(\theta ; \eta )}{\partial \eta }\bigg \}\nonumber \\ =&-E\bigg \{ \sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\frac{\partial \delta _{t-k}\delta _{t}\pi ^{-1}_{t-k}(\eta )\pi ^{-1}_{t}(\eta )U(\theta ;y_{t-k},y_{t})}{\partial \eta }\bigg \}\nonumber \\ =&E\bigg \{\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T} \frac{\delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t})\partial \pi _{t-k}(\eta )/ \partial \eta }{\pi ^{2}_{t-k}(\eta )\pi _{t}(\eta )} \nonumber \\&+ \sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\frac{ \delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t}) \partial \pi _{t}(\eta )/ \partial \eta }{\pi _{t-k}(\eta )\pi ^{2}_{t}(\eta )}\bigg \} \nonumber \\ =&E\bigg \{ \sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\frac{\delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t})(1-\pi _{t-k}(\eta ))z(x,y_{t-k};\eta )}{\pi _{t-k}(\eta )\pi _{t}(\eta )}\nonumber \\&+\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T} \frac{\delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t})(1-\pi _{t}(\eta ))z(x,y_{t};\eta )}{\pi _{t-k}(\eta )\pi _{t}(\eta )}\bigg \} \nonumber \\ =&E\bigg \{ \sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\frac{ \delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t})(\delta _{t-k}-\pi _{t-k}(\eta ))z(x,y_{t-k};\eta )}{\pi _{t-k}(\eta )\pi _{t}(\eta )} \nonumber \\&+ \sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\frac{ \delta _{t-k}\delta _{t}U(\theta ;y_{t-k},y_{t})(\delta _{t}-\pi _{t}(\eta ))z(x,y_{t};\eta )}{\pi _{t-k}(\eta )\pi _{t}(\eta )}\bigg \} \nonumber \\ =&E\Big \{\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\delta _{t-k}\delta _{t}\pi ^{-1}_{t-k}(\eta )\pi ^{-1}_{t}(\eta )U(\theta ;y_{t-k},y_{t})\nonumber \\&\cdot \big [ s^{T}(\eta ;\delta _{t-k},x,y_{t-k})+ s^{T}(\eta ;\delta _{t},x,y_{t}) \big ]\Big \}\,, \end{aligned}$$

where \(z(x,y_{t};\eta )=\pi ^{-1}_{t}(\eta )\{1-\pi _{t}(\eta )\}^{-1}\partial \pi _{t}(\eta )/\partial \eta \), yielding (4.1). \(\square \)

Since \(u(\theta ; \eta )= \sum \nolimits _{k=1}^{K}\sum \nolimits _{t=k+1}^{T}\delta _{t-k}\delta _{t}\pi ^{-1}_{t-k}(\eta )\pi ^{-1}_{t}(\eta )U(\theta ;y_{t-k},y_{t})\), and the partial derivatives \(\partial U(\theta ;y_{t-k},y_{t})/ \partial \theta ^{T}\) of \(U(\theta ; y_{t-k},y_{t})\) with respect to \(\theta \) exist for \(k=1,\ldots , K\), the equation (4.2) obviously holds.

Proof of Theorem 1

Taking expectation on \(\bar{S}(\eta )\), we have

$$\begin{aligned} E(\bar{S}(\eta ))= & {} E\Big \{\sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\big \{\delta _{i,t}s(\eta ;\delta _{i,t},x_{i},y_{i,t}) \nonumber \\&+(1-\delta _{i,t})E[s(\eta ;\delta _{i,t},x_{i},Y_{t})|x_{i},\delta _{i,t}=0]\big \}\Big \} \nonumber \\= & {} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\Big \{ E[\delta _{i,t}s(\eta ;\delta _{i,t},x_{i},y_{i,t})] \nonumber \\&+ E\{(1-\delta _{i,t})E[s(\eta ;\delta _{i,t},x_{i},Y_{t})|x_{i},\delta _{i,t}=0]\} \Big \}\nonumber \\= & {} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\Big \{ E[\delta _{i,t}\{\delta _{i,t}-\pi (x_{i},y_{i,t};\eta )\}z(x_{i},y_{i,t};\eta )]\nonumber \\&+ E\{(1-\delta _{i,t})E[\{\delta _{i,t}-\pi (x_{i},Y_{t};\eta )\}z(x_{i},Y_{t};\eta )|x_{i},\delta _{i,t}=0]\} \Big \}\nonumber \\= & {} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\Big \{ E[\pi (x_{i},y_{i,t};\eta )(1-\pi (x_{i},y_{i,t};\eta ))z(x_{i},y_{i,t};\eta )]\nonumber \\&- E\{(1-\delta _{i,t})E[(\pi (x_{i},Y_{t};\eta ))z(x_{i},Y_{t};\eta )|x_{i}]\} \Big \}\nonumber \\= & {} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\Big \{ E[\partial \pi (x_{i},y_{i,t};\eta )/\partial \eta ]\nonumber \\&- E\{E[(1-\pi (x_{i},Y_{t};\eta ))\pi (x_{i},Y_{t};\eta )z(x_{i},Y_{t};\eta )|x_{i}]\} \Big \} \nonumber \\= & {} \sum \limits _{i=1}^{n}\sum \limits _{t=2}^{T}\Big \{E[\partial \pi (x_{i},y_{i,t};\eta )/\partial \eta ]- E[\partial \pi (x_{i},Y_{t};\eta )/\partial \eta ] \Big \}=0\,. \end{aligned}$$

That is, \(\bar{S}(\eta )\) is unbiased. It follows from standard asymptotic theory that \(\widehat{\eta }\) is a consistent estimator of \(\eta _{0}\). \(\square \)

In fact, \(U_{PS}(\theta ; \widehat{\eta })\) is also unbiased because

$$\begin{aligned} E\Big \{U_{PS}(\theta ; \widehat{\eta })\Big \} =&E\Big \{\sum \limits _{i=1}^{n}\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\delta _{i,t-k}\delta _{i,t}\pi ^{-1}_{i,t-k}(\widehat{\eta })\pi ^{-1}_{i,t}(\widehat{\eta })U(\theta ;y_{i,t-k},y_{i,t}) \Big \}\nonumber \\ =&\, E\Big \{\sum \limits _{i=1}^{n}\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}E\big [\delta _{i,t-k}\delta _{i,t}\pi ^{-1}_{i,t-k}(\widehat{\eta })\pi ^{-1}_{i,t}(\widehat{\eta })\nonumber \\&\quad U(\theta ;y_{i,t-k},y_{i,t})| x_{i}, y_{i,t-k},y_{i,t}\big ] \Big \}\nonumber \\ =&\, E\Big \{\sum \limits _{i=1}^{n}\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}\pi ^{-1}_{i,t-k}(\eta _{0})\pi ^{-1}_{i,t}(\eta _{0})U(\theta ;y_{i,t-k},y_{i,t})\nonumber \\&\quad E\big [\delta _{i,t-k}\delta _{i,t}| x_{i}, y_{i,t-k},y_{i,t}\big ] \Big \}\nonumber \\ =&\, E\Big \{\sum \limits _{i=1}^{n}\sum \limits _{k=1}^{K}\sum \limits _{t=k+1}^{T}U(\theta ;y_{i,t-k},y_{i,t}) \Big \}\nonumber \\ =&\, E\Big \{\sum \limits _{k=1}^{K}U(\theta ;y_{t-k},y_{t}) \Big \}=0\,. \end{aligned}$$

Therefore, \(\widehat{\theta }_{PS}\) is a consistent estimator of \(\theta _{0}\).

Proof of Theorem 2

Let \(s(\eta )=\sum \limits _{t=2}^{T}\big \{\delta _{t}s(\eta ;\delta _{t},x,y_{t})+(1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\big \}\) and

$$\begin{aligned}&\dot{s}(\eta ;\delta _{t},x,y_{t}) = \partial s(\eta ;\delta _{t},x,y_{t})/ \partial \eta ^{T} \nonumber \\&\quad = (\delta _{t}-\pi (x,y_{t};\eta ))\frac{ \partial z(x,y_{t};\eta )}{\partial \eta ^{T}}- z(x,y_{t};\eta )\frac{ \partial \pi (x,y_{t};\eta )}{\partial \eta ^{T}}\,. \end{aligned}$$

We have

$$\begin{aligned} \frac{\partial s(\eta )}{\partial \eta ^{T}} = \sum \limits _{t=2}^{T}\bigg \{\delta _{t} \dot{s}(\eta ;\delta _{t},x,y_{t}) + (1-\delta _{t}) \frac{\partial E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]}{\partial \eta ^{T}}\bigg \}\,. \end{aligned}$$


$$\begin{aligned} E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]=\frac{\int s(\eta ;\delta _{t},x,y)f(y|x,\delta _{t}=1)O(x,y;\eta )dy}{\int O(x,y;\eta )f(y|x,\delta _{t}=1)dy}\,, \end{aligned}$$

the derivative \(\partial E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]/ \partial \eta ^{T}\) in the Eq. (A8) can be written as

$$\begin{aligned} \frac{\partial E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]}{\partial \eta ^{T}}=&\frac{\int \dot{s}(\eta ;\delta _{t},x,y)\hat{f}(y|x,\delta _{t}=1)O(x,y;\eta )dy}{\int O(x,y;\eta )\hat{f}(y|x,\delta _{t}=1)dy}\nonumber \\&+ \frac{\int s(\eta ;\delta _{t},x,y)\hat{f}(y|x,\delta _{t}=1) \partial O(x,y;\eta )/ \partial \eta ^{T} dy}{\int O(x,y;\eta )\hat{f}(y|x,\delta _{t}=1)dy}\nonumber \\&- \frac{\int s(\eta ;\delta _{t},x,y)\hat{f}(y|x,\delta _{t}=1)O(x,y;\eta )dy}{\int O(x,y;\eta )\hat{f}(y|x,\delta _{t}=1)dy} \nonumber \\&\cdot \frac{\int \hat{f}(y|x,\delta _{t}=1)\partial O(x,y;\eta )/ \partial \eta ^{T} dy}{\int O(x,y;\eta )\hat{f}(y|x,\delta _{t}=1)dy}\,. \end{aligned}$$

By \(\partial O(x,y_{t};\eta )/ \partial \eta ^{T} = - (\partial \pi (x,y_{t};\eta )/ \partial \eta ^{T}) / \pi ^{2}(x,y_{t};\eta ) = - O(x,y_{t};\eta )z(x,y_{t};\eta )\), we obtain

$$\begin{aligned}&\frac{\partial E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]}{\partial \eta ^{T}} = E[\dot{s}(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0] \nonumber \\&\quad - E[s(\eta ;\delta _{t},x,Y_{t})z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0]\nonumber \\&\quad + E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\cdot E[z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \,. \end{aligned}$$

Combining equations (A8) and (A11), we have

$$\begin{aligned} E\bigg (\frac{\partial s(\eta )}{\partial \eta ^{T}} \bigg )=&E \bigg \{\sum \limits _{t=2}^{T}\Big \{ \delta _{t} \dot{s}(\eta ;\delta _{t},x,y_{t}) + (1-\delta _{t})E[\dot{s}(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\nonumber \\&- (1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \nonumber \\&+ (1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\cdot E[z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \Big \}\bigg \}\nonumber \\ =&\, E \Big \{ \sum \limits _{t=2}^{T}\big \{\delta _{t} \dot{s}(\eta ;\delta _{t},x,y_{t}) + (1-\delta _{t})E[\dot{s}(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\big \}\Big \}\nonumber \\&- E \Big \{\sum \limits _{t=2}^{T}(1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \Big \}\nonumber \\&+ E \Big \{\sum \limits _{t=2}^{T}(1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\cdot E[z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \Big \} \,. \end{aligned}$$


$$\begin{aligned}&E \Big \{ \sum \limits _{t=2}^{T}\big \{\delta _{t} \dot{s}(\eta ;\delta _{t},x,y_{t}) + (1-\delta _{t})E[\dot{s}(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\big \}\Big \} \nonumber \\&\quad = E \Big \{ \sum \limits _{t=2}^{T}\big \{ \pi (x,y_{t};\eta ) \dot{s}(\eta ;\delta _{t},x,y_{t}) + (1-\pi (x,y_{t};\eta ))\dot{s}(\eta ;\delta _{t},x,y_{t})\big \}\Big \}\nonumber \\&\quad = E \Big [\sum \limits _{t=2}^{T}\dot{s}(\eta ;\delta _{t},x,y_{t})\Big ] \end{aligned}$$


$$\begin{aligned}&E \Big \{\sum \limits _{t=2}^{T}(1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0]\Big \}\nonumber \\&\quad = - E \Big \{\sum \limits _{t=2}^{T}(1-\delta _{t})E[\pi (x,Y_{t};\eta )z(x,Y_{t};\eta ) z^{T}(x,Y_{t};\eta )|x]\Big \}\nonumber \\&\quad = - E \Big \{E\big [\sum \limits _{t=2}^{T}(1-\pi (x,Y_{t};\eta ))\pi (x,Y_{t};\eta )z(x,Y_{t};\eta ) z^{T}(x,Y_{t};\eta )|x\big ]\Big \}\nonumber \\&\quad = - E\Big [\sum \limits _{t=2}^{T}z(x,Y_{t};\eta ) \partial \pi (x,y_{t};\eta ))/ \partial \eta ^{T}\Big ]\nonumber \\&\quad = E \Big [\sum \limits _{t=2}^{T}\dot{s}(\eta ;\delta _{t},x,y_{t})\Big ] \,, \end{aligned}$$

we then have

$$\begin{aligned} E\bigg (\frac{\partial s(\eta )}{\partial \eta ^{T}} \bigg ) =&E \Big \{\sum \limits _{t=2}^{T}(1-\delta _{t})E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\cdot E[z^{T}(x,Y_{t};\eta ) |x,\delta _{t}=0] \Big \} \,. \end{aligned}$$

Let \(\mathcal {I}_{11}(\eta )= - E\big (\partial s(\eta )/ \partial \eta ^{T} \big )\). Since \(\widehat{\eta }\) is the solution to the mean score equation \(\bar{S}(\eta )=0\) in (3.3), we have

$$\begin{aligned} \sqrt{n}\left( \widehat{\eta }-\eta _{0} \right) \xrightarrow {\ d\ }N\left( 0, \mathcal {I}^{-1}_{11}(\eta _{0}) \right) \,, \end{aligned}$$


$$\begin{aligned} \mathcal {I}^{-1}_{11}(\eta _{0}) =&-\bigg \{E\Big [\sum \limits _{t=2}^{T}(1-\delta _{t})E[s(\eta _{0};\delta _{t},x,Y_{t})|x,\delta _{t}=0]E[z^{T}(x,Y_{t};\eta _{0})|x,\delta _{t}=0]\Big ]\bigg \}^{-1} \nonumber \\ =&-\bigg \{E\Big [\sum \limits _{t=2}^{T}(1-\delta _{t})\bar{s}_{0}(\eta _0;x)\bar{z}_{0}(\eta _0;x)\Big ]\bigg \}^{-1} \,, \end{aligned}$$

\(\bar{s}_{0}(\eta ;x)=E[s(\eta ;\delta _{t},x,Y_{t})|x,\delta _{t}=0]\) and \(\bar{z}_{0}(\eta ;x)=E[z^{T}(x,Y_{t};\eta )|x,\delta _{t}=0]\), completing the proof of Theorem 2. \(\square \)

Proof of Theorem 3

Since \(\widehat{\psi } = (\widehat{\eta }, \widehat{\theta }_{PS})\) is the solution to

$$\begin{aligned} \left( \begin{array}{c} \bar{S}(\eta )=0\\ U_{PS}(\theta ; \eta )=0 \end{array}\right) =\left( \begin{array}{c} 0\\ 0 \end{array} \right) \,, \end{aligned}$$

the variance of \(\widehat{\psi } = (\widehat{\eta }, \widehat{\theta }_{PS})\) can be obtained by

$$\begin{aligned} \mathrm{var}\left\{ \sqrt{n}\left( \begin{array}{c} \widehat{\eta }-\eta _{0}\\ \widehat{\theta }_{PS}-\theta _{0} \end{array}\right) \right\} = \mathcal {I}^{-1} \mathrm{var}\left\{ \begin{array}{c} s(\eta _{0})\\ u(\theta _{0}; \eta _{0}) \end{array}\right\} \mathcal {I}^{-T} \,, \end{aligned}$$

where \(\mathcal {I}=\mathcal {I}(\theta _{0}, \eta _{0})\) and

$$\begin{aligned} \mathcal {I}(\theta , \eta ) = \left( \begin{array}{cc} \mathcal {I}_{11}(\eta )&{} 0\\ \mathcal {I}_{21}(\theta ,\eta )&{} \mathcal {I}_{22}(\theta ,\eta ) \end{array}\right) = -\left\{ \begin{array}{cc} E\big (\frac{\partial s(\eta )}{\partial \eta ^{T}}\big )&{} E\big (\frac{\partial s(\eta )}{\partial \theta ^{T}}\big )\\ E\big (\frac{\partial u(\theta ; \eta )}{\partial \eta ^{T}}\big ) &{} E\big (\frac{\partial u(\theta ; \eta )}{\partial \theta ^{T}}\big ) \end{array}\right\} \,. \end{aligned}$$

Further, we have

$$\begin{aligned} \mathcal {I}^{-1}= \left( \begin{array}{cc} \mathcal {I}^{-1}_{11}&{} 0\\ -\mathcal {I}^{-1}_{22}\mathcal {I}_{21}\mathcal {I}^{-1}_{11}&{} \mathcal {I}^{-1}_{22} \end{array}\right) \,, \end{aligned}$$

where \(\mathcal {I}_{11} = \mathcal {I}_{11}(\eta _{0})\), \(\mathcal {I}_{22} = \mathcal {I}_{22}(\theta _{0}, \eta _{0})\), \(\mathcal {I}_{21} = \mathcal {I}_{21}(\theta _{0}, \eta _{0})\). Combining equations (A19) and (A21), we have

$$\begin{aligned} \Sigma _{\theta } =&\mathrm{var}(\sqrt{n}(\widehat{\theta }_{PS}-\theta _{0}))= \mathrm{var}\big \{\mathcal {I}^{-1}_{22}[u(\theta _{0}; \eta _{0})-\mathcal {I}_{21}\mathcal {I}^{-1}_{11}s(\eta _{0})]\big \}\nonumber \\ =&\mathcal {I}^{-1}_{22} \mathrm{var}\big [u(\theta _{0}; \eta _{0})-\mathcal {I}_{21}\mathcal {I}^{-1}_{11}s(\eta _{0})\big ] \mathcal {I}^{-T}_{22} \,. \end{aligned}$$

It follows that

$$\begin{aligned} \sqrt{n}\left( \widehat{\theta }_{PS}-\theta _{0} \right) \xrightarrow {\ d\ }N\left( 0, \Sigma _{\theta } \right) \,, \end{aligned}$$

as \(n \rightarrow \infty \), where

$$\begin{aligned} \Sigma _{\theta }= \mathcal {I}^{-1}_{22} \mathrm{var}[u(\theta _{0}; \eta _{0})- \mathcal {I}_{21}\mathcal {I}^{-1}_{11}s(\eta _{0})]\mathcal {I}^{-T}_{22}\,. \end{aligned}$$

That is, Theorem 3 holds. \(\square \)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Liu, Z., Yau, C.Y. A propensity score adjustment method for longitudinal time series models under nonignorable nonresponse. Stat Papers (2021).

Download citation


  • Composite likelihood
  • Consecutive pairwise likelihood
  • Estimating equation
  • Missing not at random