Abstract
In real data analysis, practitioners frequently come across the case that a discrete response will be related to both a function-valued random variable and a vector-value random variable as the predictor variables. In this paper, we consider the generalized functional partially linear models (GFPLM). The infinite slope function in the GFPLM is estimated by the principal component basis function approximations. Then, we consider the theoretical properties of the estimator obtained by maximizing the quasi likelihood function. The asymptotic normality of the estimator of the finite dimensional parameter and the rate of convergence of the estimator of the infinite dimensional slope function are established, respectively. We investigate the finite sample properties of the estimation procedure via Monte Carlo simulation studies and a real data analysis.
Similar content being viewed by others
References
Aneiros-Pérez G, Vieu P (2006) Semi-functional partial linear regression. Stat Probab Lett 76(11):1102–1110
Aneiros-Pérez G, Vieu P (2015) Partial linear modelling with multi-functional covariates. Comput Stat 30(3):647–671
Cai T, Hall P (2006) Prediction in functional linear regression. Ann Stat 34(5):2159–2179
Cai T, Yuan M (2012) Minimax and adaptive prediction for functional linear regression. J Am Stat Assoc 107(499):1201–1216
Cardot H, Sarda P (2005) Estimation in generalized linear model for functional data via penalized likelihood. J Multivar Anal 92(1):24–41
Cardot H, Ferraty F, Sarda P (1999) Functional linear model. Stat Probab Lett 45(1):11–22
Crambes C, Kneip A, Sarda P (2009) Smoothing splines estimators for functional linear regression. Ann Stat 37(1):35–72
Febrero-Bande M, Oviedo de la Fuente M (2012) fda.usc: functional data analysis. Utilities for Statistical Computing. http://cran.r-project.org/web/packages/fda.usc/index.html. R package version 1.3.0
Ferraty F, Vieu P (2006) Nonparametric functional data analysis. Springer, New York
Geyer CJ (1994) On the convergence of Monte Carlo maximum likelihood calculations. J R Stat Soc B 56(1):261–274
Hall P, Horowitz J (2007) Methodology and convergence rates for functional linear regression. Ann Stat 35:70–91
Horváth L, Kokoszka P (2012) Inference for functional data with applications. Springer, New York
Kong D, Staicu A-M, Maity A (2016a) Classical testing in functional linear models. J Nonparametric Stat 28:813–838
Kong D, Xue K, Yao F, Zhang HH (2016b) Partially functional linear regression in high dimensions. Biometrika 103:147–159
Kong D, Ibrahim JG, Lee E, Zhu H (2018) FLCRM: functional linear cox regression model. Biometrics 74:109–117
Li Y, Hsing T (2007) On rates of convergence in functional linear regression. J Multivariate Anal 98:1782–1804
Ling N, Aneiros G, Vieu P (2018) kNN estimation in functional partial linear modeling. Stat Pap (in press)
Lu Y, Du J, Sun Z (2014) Functional partially linear quantile regression model. Metrika 77(2):317–332
Mccullagh P, Nelder JA (1989) An outline of generalized linear models. Springer, New York
Müller H, Stadtmüller U (2005) Generalized functional linear models. Ann Stat 33:774–805
Nelder JA, Wedderburn RWM (1972) Generalized linear models. J R Stat Soc B 135(3):370–384
Peng Q, Zhou J, Tang N (2016) Varying coefficient partially functional linear regression models. Stat Pap 57:827–841
Pollard D (1991) Asymptotics for least absolute deviation regression estimators. Econom Theory 7(2):186–199
Ramsay JO, Dalzell CJ (1991) Some tools for functional data analysis. J R Stat Soc 53(3):539–572
Ramsay JO, Silverman BW (2005) Functional data analysis, 2nd edn. Springer, New York
Riesz F, Sz.-Nagy B (1956) Vorlesungen über funktionalanalysis. Veb Deutscher Verlag Der Wissenschaften Berlin
Shin H (2009) Partial functional linear regression. J Stat Plan Inference 139:3405–3418
van der Vaart AW (1998) Asymptotic Statistics. Cambridge University Press, Cambridge
Wang L, Liu X, Liang H, Carroll RJ (2011) Estimation and variable selection for generalized additive partial linear models. Ann Stat 39(4):1827–1851
Yao F, Müller H, Wang J (2005) Functional linear regression analysis for longitudinal data. Ann Stat 33:2873–2903
Yu P, Zhang Z, Du J (2016) A test of linearity in partial functional linear regression. Metrika 79(8):1–17
Yu P, Du J, Zhang Z (2017) Varying-coefficient partially functional linear quantile regression models. J Korean Stat Soc 46:462–475
Yu P, Du J, Zhang Z (2018) Single-index partially functional linear regression model. Stat Pap (in press)
Zhang X, Hua L (2011) Focused information criterion and model averaging for generalized additive partial linear models. Ann Stat 39(1):174–200
Zhou J, Chen M (2012) Spline estimators for semi-functional linear model. Stat Probab Lett 82:505–513
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Du’s work is supported by the National Natural Science Foundation of China (Nos. 11501018, 11771032), the Science and Technology Project of Beijing Municipal Education Commission (KM201910005015) and Program for Rixin Talents in Beijing University of Technology. Cao’s work is supported by the National Natural Science Foundation of China (No. 11701020). Zhou’s work is supported by the National Natural Science Foundation of China (No. 11861074). Xie’s work is supported by the National Natural Science Foundation of China (No. 11571340) and the Science and Technology Project of Beijing Municipal Education Commission (KM201710005032).
Appendix: Proofs
Appendix: Proofs
Before we present the proof of Theorem 1, we give some useful lemmas and their proofs. For any matrix \(\varvec{A}\), denote \(L_2\) norm for \(\varvec{A}\) as \(\Vert \varvec{A}\Vert _2 = \sup \limits _{\Vert \varvec{x}\Vert \ne 0}\frac{ \Vert \varvec{A}\varvec{x}\Vert }{\Vert \varvec{x}\Vert }.\) Let \(\eta _{0i}=\varvec{Z}_i^T\varvec{\alpha }_0+\int _0^1X_i(t)\beta _0(t)dt,\) and \(\tilde{\eta }_i=\varvec{Z}_i^T{\varvec{\alpha }}_0+\varvec{U}_i^T\varvec{\gamma }_0.\)
Lemma 1
Let \(R_i=\int _0^1X_i(t)\beta _0(t)dt-\varvec{U}_i^T\varvec{\gamma }_0\), for \(i=1,2,\dots ,n\). Under conditions C1–C7, it has
Furthermore, one has \(\eta _{0i}-\tilde{\eta }_i=O_p\left( n^{-(2b+a-1)/2(a+2b)}\right) .\)
Proof of Lemma 1
By the Karhunen–Loève representation and the fact that sequence \(\{\phi _j\}\) forms an orthonormal basis in \(L^2([0, 1])\), we have
where
and
First, we consider \(I_1.\) By triangle inequality, C1 and C2, we have
Next, we consider \(I_2\). Invoking the fact that \(x_{ik}=\langle X_i,\phi _k\rangle \) are uncorrelated random variables, we have
By C3, we have \(I_2=O_p\left( m^{-(2b+a-1)/2}\right) =O_p\left( n^{-\frac{(2b+a-1)}{2(a+2b)}}\right) .\) Combining the convergence rate of \(I_1\) with \(I_2\)’s, we have
By direct calculation and the fact \(\Vert R_i\Vert ^2=\Vert \eta _{0i}-\tilde{\eta }_i \Vert ^2=O_p\left( n^{-(2b+a-1)/(a+2b)}\right) \), we have
Thus Lemma 1 is proved. \(\square \)
Let
where \( \tilde{ \varvec{Z}}_i =\varvec{Z}_i- {\varvec{h}}(X_i)\) and \(\varvec{\gamma }_0=( \langle \beta _0, \hat{\phi }_1\rangle ,\langle \beta _0, \hat{\phi }_2\rangle ,\dots ,\langle \beta _0, \hat{\phi }_m\rangle )^T.\) The following lemma gives the asymptotic distribution of \(\check{\varvec{\alpha }}.\)
Lemma 2
Under conditions C1–C7, it has
where \( \Sigma _{1} =E[\rho _2(\eta _0)\tilde{\varvec{Z}}\tilde{ \varvec{Z}}^T]\), \( \Sigma _2 =E[q_1^2(\eta _0)\tilde{\varvec{Z}} \tilde{\varvec{Z}}^T]\) and \(\tilde{ \varvec{Z}}=\varvec{Z}-\varvec{h}(X)\).
Proof of Lemma 2
Let \(\varvec{u}=\sqrt{n}( \varvec{\alpha }- {\varvec{\alpha }}_0) \) and \(\check{\varvec{u}}= \sqrt{n}(\check{\varvec{\alpha }}- {\varvec{\alpha }}_0).\) Since that \(\check{\varvec{\alpha }}\) maximizes \( \sum _{i=1}^n Q\left( Y_i,g^{-1}\left( \tilde{ \varvec{Z}}_i^T\varvec{\alpha }+\varvec{U}_i^T\varvec{\gamma }_0+{\varvec{h}}(X_i)\varvec{\alpha }_0\right) \right) \), we have \(\check{\varvec{u}}\) maximizes
Invoking Taylor expansion, one has
where
with \(\zeta _i\) lies between 0 and \(\tilde{ \varvec{Z}}_i^T\varvec{u}/\sqrt{n}\) for \(i=1,2,\dots ,n.\)
Denote \( \mathcal {F}_n=\{ q_2(y,u): y\in R, | u |\le C\}, \) where C is a constant. By Theorem 2.7.1 of van der Vaart and Wellner (1996), one has that \(\mathcal {F}_n\) is a P-Donsker class of measurable functions. Thus, by Theorem 19.24 of van der Vaart (1998), we have
In addition,
where
and
By routine calculation, we have \(I_{4}= O_p\left( n^{-(2b+a-1)/2(a+2b)}\right) =o_p(1).\)
Therefore, we have
Let \(\varvec{W}_n=\frac{1}{\sqrt{n}}\sum _{i=1}^n q_1\left( Y_i, \eta _{0i} \right) \tilde{\varvec{Z}}_i.\) By the Lindeberg-Feller central limit theorem, we have
where \( \Sigma _2 =E[q_1^2(Y,\eta _0)\tilde{\varvec{Z}} \tilde{\varvec{Z}}^T].\) Combining (5) with (7), we have
Here, the epi-convergence results of Geyer (1994) and Convexity Lemma of Pollard (1991) imply that
By Slutsky Lemma, we have
Thus, the proof for Lemma 2 is completed. \(\square \)
Denote \(\hat{\varvec{\theta }}=(\hat{\varvec{\alpha }}^T,\hat{\varvec{\gamma }}^T)^T\) and \( \check{{\varvec{\theta }}}=( {\check{\varvec{\alpha }}}^T, {\varvec{\gamma }_0}^T)^T.\) We further have the following lemma.
Lemma 3
Under conditions C1–C7, it has
Proof of Lemma 3
Note that
where \(\bar{\varvec{\theta }}=\nu \hat{\varvec{\theta }}+(1-\nu )\check{{\varvec{\theta }}}\), and \(0\le \nu \le 1.\) By the fact that \( \frac{\partial L(\hat{\varvec{\theta }})}{\partial \varvec{\theta }}=0,\) we have
First, we consider \(\frac{\partial L(\varvec{\theta })}{\partial \varvec{\theta }} \mid _{\varvec{\theta }= \check{{\varvec{\theta }}}}.\)
where \( \frac{\partial L(\check{{\varvec{\theta }}})}{\partial \varvec{\alpha }}= \sum _{i=1}^n q_1\left( Y_i, \check{\eta }_{i}\right) \varvec{Z}_i \), \(\frac{\partial L(\check{{\varvec{\theta }}})}{\partial \varvec{\gamma }} = \sum _{i=1}^n q_1\left( Y_i, \check{\eta }_{i}\right) \varvec{U}_i. \) and \(\check{{\eta _i}}=\varvec{Z}_i^T{\check{\alpha }}+\varvec{U}_i^T\varvec{\gamma }_0,\ i=1,\ldots ,n\). Note that
and
where \({ \eta }^*_{i}=\nu \eta _{0i}+(1-\nu )\check{\eta }_{i},\) \(0\le \nu \le 1\) for \(i=1,2,\dots ,n.\) By routine calculation, we have
In addition, by condition C2 and the fact \(\Vert R_i\Vert ^2=O_p(n^{-(2b+a-1)/(a+2b)})\), we have
Combining (9) with (10), one has \( \frac{\partial L(\check{{\varvec{\theta }}})}{\partial \varvec{\alpha }}= \sum _{i=1}^n q_1\left( Y_i, \check{\eta }_{i}\right) \varvec{Z}_i=O_p\left( n n^{-(2b+a-1)/2(a+2b) }\right) . \)
Similarly, we have
and
Combining (11) with (12), one has \( \frac{\partial L(\check{{\varvec{\theta }}})}{\partial \varvec{\gamma } }= \sum _{i=1}^n q_1\left( Y_i, {\eta }^*_{i}\right) \varvec{U}_i= O_p\left( n n^{-(2b+a-1)/2(a+2b) }\right) . \) Therefore, we have
Following Lemma A.3 of Wang et al. (2011), we have
\(\square \)
Proof of Theorem 1
Similar to Lu et al. (2014), one has
Now, we consider \(I_{5}.\) Note that \(\Vert \phi _{j}-\hat{\phi }_{j}\Vert ^2=O_p(n^{-1}j^2)\), one has
Next, we consider \(I_{6}.\)
Therefore, combining these with Lemma 3, we have
Next, we prove the asymptotic normality of \(\hat{\varvec{\alpha }}\). Let \(\hat{\eta }_{i}=\varvec{Z}_i^T\hat{\varvec{\alpha }}+\varvec{U}_i^T\hat{\varvec{\gamma }}.\) For any \(\varvec{v}\in R^p\), define
Note that when \(\varvec{v} = 0,\) \(\hat{\eta }(\varvec{v})\) maximizes \(L(\hat{\eta }(\varvec{v}))= \sum _{i=1}^nQ\left( Y_i,g^{-1}\left( \hat{\eta }_i(\varvec{v})\right) \right) .\) Thus, one has
By Lemma 3, we have
We first consider \(A_2.\) Obviously, \(E(A_2)=0.\) By Lemma 1, we have
Thus, one has \(A_2=O_p\left( \sqrt{m}\right) =o_p\left( {\sqrt{n}}\right) .\) Now, we consider \(A_3.\) Note that
Invoking Lemmas 2 and 3, we have
Thus, by \(A_1,\) \(A_2\) and \(A_3\), we have
Furthermore, by the central limit theorem and slutsky’s Lemma, we have
where \(\Sigma _{1}=E\left( \rho _2( {\eta _0})\tilde{ \varvec{Z}}\tilde{ \varvec{Z}}^T\right) \) and \(\Sigma _{2}=E\left( q_1^2( {\eta _0})\tilde{ \varvec{Z}}\tilde{\varvec{Z}}^T\right) \). \(\square \)
Rights and permissions
About this article
Cite this article
Cao, R., Du, J., Zhou, J. et al. FPCA-based estimation for generalized functional partially linear models. Stat Papers 61, 2715–2735 (2020). https://doi.org/10.1007/s00362-018-01066-8
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00362-018-01066-8
Keywords
- Generalized linear model
- Functional partially linear model
- Quasi likelihood
- Karhunen–Loève representation