Abstract
This paper studies a partially linear additive regression with spatial data. A new estimation procedure is developed for estimating the unknown parameters and additive components in regression. The proposed method is suitable for high dimensional data, there is no need to solve the restricted minimization problem and no iterative algorithms are needed. Under mild regularity assumptions, the asymptotic distribution of the estimator of the unknown parameter vector is established, the asymptotic distributions of the estimators of the unknown functions are also derived. Finite sample properties of our procedures are studied through Monte Carlo simulations. A real data example about spatial soil data is used to illustrate our proposed methodology.
Similar content being viewed by others
References
Al-Sulami D, Jiang Z, Lu Z, Zhu J (2017) Estimation for semiparametric nonlinear regression of irregularly located spatial time-series data. Econom Stat 2:22–35
Athreya KB, Pantala SG (1986) A note on strong mixing of ARMA processes. Stat Probab Lett 4:187–190
Brieman L, Friedman J (1985) Estimating optimal transformations for multiple regression and correlation (with discussion). J Am Stat Assoc 80:580–619
Buja A, Hastie T, Tibshirani R (1989) Linear smoothers and additive models (with discussion). Ann Stat 17:453–555
Cheng G, Zhou L, Huang J (2014) Efficient semiparametric estimation in generalized partially linear additive models for longitudinal/clustered data. Bernoulli 20:141–163
Cressie NAC (1991) Statistics for spatial data. Wiley, New York
Fan Y, Li Q (2003) A kernel-based method for estimating additive partially linear models. Stat Sin 13:739–62
Friedman J, Stuetzle W (1981) Projection pursuit regression. J Am Stat Assoc 76:817–823
Gao J, Lu Z, Tjøstheim D (2006) Estimation in semiparametric spatial regression. Ann Stat 34:1395–1435
Hallin M, Lu Z, Tran LT (2001) Density estimation for spatial linear processes. Bernoulli 7:657–668
Hallin M, Lu Z, Tran LT (2004) Local linear spatial regression. Ann Stat 32:2469–2500
Huang J, Horowitz J, Wei F (2010) Variable selection in nonparametric additive models. Ann Stat 38:2282–2313
Lee YK, Choi H, Park BU, Yu KS (2004) Local likelihood density estimation on random fields. Stat Probab Lett 68:347–57
Li Q (2000) Efficient estimation of additive partially linear models. Int Econ Rev 41:1073–1092
Lian H, Liang H, Ruppert D (2015) Separation of covariates into nonparametric and parametric parts in high-dimensional partially linear additive models. Stat Sin 25:591–607
Liang H, Thurston S, Ruppert D, Apanasovich T, Hauser R (2008) Additive partial linear models with measurement errors. Biometrika 95:667–678
Nandy S, Lim C, Maiti T (2017) Additive model building for spatial regression. J R Stat Soc B 79:779–800
Ramsay T (2002) Spline smoothing over difficult regions. J R Stat Soc B 64:307–319
Ripley B (1981) Spatial statistics. Wiley, New York
Rosenblatt M (1956) A central limit theorem and a strong mixing condition. Proc Natl Acad Sci USA 42:43–47
Sherwood B, Wang L (2016) Partially linear additive quantile regression in ultra-high dimension. Ann Stat 44:288–317
Stone C (1985) Additive regression and other nonparametric models. Ann Stat 13:689–705
Tang Q, Cheng L (2009) B-spline estimation for varying coefficient regression with spatial data. Sci Chin Ser A 52(11):2321–2340
Tran LT (1990) Kernel density estimation on random field. J Multivar Anal 34:37–53
Yang J, Yang H, Lu F (2019) Rank-based shrinkage estimation for identification in semiparametric additive models. Stat Pap 60:1255–1281
You J, Zhou X (2013) Efficient estimation in panel data partially additive linear model with serially correlated errors. Stat Sin 23:271–303
Yu S, Wang G, Wang L, Liu C, Yang L (2020) Estimation and Inference for Generalized Geoadditive Models. J Am Stat Assoc 115:761–774
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.
Appendix: Proofs
Appendix: Proofs
In this section, let \(C>0\) denote a generic constant of which the value may change from line to line. For a matrix \(B=(b_{ij})\), set \(\Vert B\Vert _{\infty }=\max _{i}\sum _{j}|b_{ij}|\) and \(|B|_{\infty }=\max _{i,j}|b_{ij}|\). For a vector \(v=(v_{1},\ldots ,v_{k})^{T}\), set \(\Vert v\Vert _{\infty }=\sum _{j=1}^{k}|v_{j}|\) and \(|v|_{\infty }=\max _{1\le j\le k}|v_{j}|\).
Let \(f_{r\nu }(x_{r})=\chi _{r\nu }(x_{r})f_{r}(x_{r})\), \({\bar{f}}_{r\nu }=(\sum _{i=1}^{m}\sum _{j=1}^{n}f_{r\nu }(X_{ijr})) /(\sum _{i=1}^{m}\sum _{j=1}^{n}\chi _{r\nu }(X_{ijr}))\),
\(\bar{{\check{f}}}_{r\nu }=(\sum _{i=1}^{m}\sum _{j=1}^{n}{\check{f}}_{r\nu }(X_{ijr})) /(\sum _{i=1}^{m}\sum _{j=1}^{n}\chi _{r\nu }(X_{ijr}))\) and \(f_{r\nu }^{*}(X_{ijr})=[f_{r\nu }(X_{ijr})-\chi _{r\nu }(X_{ijr}){\bar{f}}_{r\nu }] -[{\check{f}}_{r\nu }(X_{ijr})-\chi _{r\nu }(X_{ijr})\bar{{\check{f}}}_{r\nu }]\). Noting that \(f_{r}(x_{r})=\sum _{\nu =1}^{M_{rN}}f_{r\nu }(x_{r})\) and \(f_{r\nu }(X_{ijr})=\chi _{r\nu }(X_{ijr}){\bar{f}}_{r\nu }+[{\check{f}}_{r\nu }(X_{ijr}) -\chi _{r\nu }(X_{ijr})\bar{{\check{f}}}_{r\nu }]+f_{r\nu }^{*}(X_{ijr})\), we get that
where \(\pmb {a}_{0r}=(\pmb {a}_{0r1}^{T},\cdots ,\pmb {a}_{0rM_{rN}}^{T})^{T}\) with \(\pmb {a}_{0r\nu }=({\bar{f}}_{r\nu },h_{r0}f_{r}'(x_{r\nu }),\ldots ,h_{r0}^{p_{r}}f_{r}^{(p_{r})} (x_{r\nu })/p_{r}!)^{T}\) for \(\nu =1,\ldots ,M_{rN}-1\) and \(\pmb {a}_{0rM_{rN}}=(h_{r0}f_{r}'(x_{rM_{rN}}),\ldots ,h_{r0}^{p_{r}}f_{r}^{(p_{r})}(x_{rM_{rN}})/p_{r}!)^{T}\), \({\bar{F}}_{rM_{rN}}=(\sum _{i=1}^{m}\sum _{j=1}^{n}f_{r}(X_{ijr}))/(\sum _{i=1}^{m}\sum _{j=1}^{n}\chi _{rM_{rN}}(X_{ijr}))\). Let \(f_{kr\nu }(x_{r})=\chi _{r\nu }(x_{r})f_{kr}(x_{r})\), \(f_{kr\nu }^{*}(X_{ijr}), k=1,\ldots ,d_{1}\) and \({\bar{F}}_{krM_{rN}}\) are defined similarly as \(f_{r\nu }^{*}(X_{ijr})\) and \({\bar{F}}_{rM_{rN}}\). Denote \(\vec {Y}={\bar{Y}}-E(Y)\), \(\vec {\pmb {Z}}=\bar{\pmb {Z}}-E(\pmb {Z})\), \(\breve{\pmb {Z}}_{ij}=(\breve{Z}_{ij1},\ldots ,\breve{Z}_{ijd_{1}})^{T}\) with
Let
Then, we have
with
Lemma A.1
Under assumptions 1–5, it holds that
Proof
Let \(D_{N}=\{(i,j): 1\le i\le m, 1\le j\le n\}\), \({\tilde{A}}_{r\nu k}(X_{ijr})=\breve{A}_{r\nu k}(X_{ijr})-E(\breve{A}_{r\nu k}(X_{ijr}))\) and \(A^{*}=\max _{1\le r\le d_{2}, 1\le \nu \le M_{rN}, 1\le k\le p_{r}}\big |\frac{1}{N}\sum _{i=1}^{m}\sum _{j=1}^{n}{\tilde{A}}_{r\nu k}(X_{ijr})|\). For any sufficiently small positive constant \(\varepsilon \), we have
Let \(c_{Nk}=[M_{N}^{\delta /((2+\delta )\tau )}]\) for \(k=1,2\), where \(\tau >2(4+\delta )/(2+\delta )\) is a constant. Let the set \(\{(i,j)\ne (i',j')\in D_{N}\}\) be split into the following two parts
By assumption 5, we have
Turning to \({\mathbf {S}}_{2}\), using Lemma 5.1 of Hallin et al. (2004b), we obtain that
Therefore, by assumptions 4 and 5 , we get
Now Lemma A.1 follows from (A.6)–(A.8) and the fact that \(\sum _{(i,j)\in D_{N}}E(\breve{A}_{r\nu k}^{2}(X_{ijr}))\le NM_{N}^{-1}\). \(\square \)
Lemma A.2
Under assumptions 1-5, it holds that
Proof
We first prove
Let \(\xi _{ijr\nu 1}=\chi _{r\nu }(X_{ijr})-E(\chi _{r\nu }(X_{r}))\chi _{rM_{rN}}(X_{ijr}) /E(\chi _{rM_{rN}}(X_{r}))\),
Then
Similar to the proof of Lemma A.1, we obtain that \(E[\sum _{i=1}^{m}\sum _{j=1}^{n}\xi _{ijr\nu 1}V_{ijl}]^{2}\le CNM_{N}^{2\delta /((2+\delta )\tau )-1}\) and \(E[\sum _{i=1}^{m}\sum _{j=1}^{n}\chi _{r\nu }(X_{ijr})\eta _{ijr\nu k1}V_{ijl}]^{2}\le CNM_{N}^{2\delta /((2+\delta )\tau )-1}\). Hence,
Lemma A.1 implies \(\max _{1\le r\le d_{2}, 1\le \nu \le M_{rN}-1}|\xi _{r\nu 2}|=o_{p}(M_{N}^{-1})\), \(\max _{1\le r\le d_{2}, 1\le \nu \le M_{rN}}|\eta _{r\nu k2}|=o_{p}(M_{N}^{-1})\). Since \(E[\chi _{r\nu }(X_{ijr})V_{ijl}]=E[(\chi _{r\nu }(X_{ijr})-E\chi _{r\nu }(X_{ijr}))V_{ijl}] +E(\chi _{r\nu }(X_{ijr}))E(V_{ijl})=0\), then by arguments similar to those used in the proof of Lemma A.1, we have
Therefore,
Now (A.9) follows from (A.10)–(A.12), (A.14) and (A.15). Similar to the proof of (A.9), we deduce that
Using the fact that \(E(f_{r}(X_{r}))=0\), we get that \({\bar{F}}_{rM_{rN}}=O_{p}(M_{N}^{3/2}/N^{1/2})\). Hence,
Similar to the proof of (A.9) and using Assumptions , we have
Similar to the proof of (A.17), we get that
Now Lemma A.2 follows from (A.9) and (A.17)–(A.19)and Assumption 5. \(\square \)
Lemma A.3
Under Assumptions 1-5, it holds that
Proof
We first prove that \(M_{N}\pmb {A}_{N}/N\) is invertible. Let \(\lambda _{min}\) be the minimum eigenvalue of \(M_{N}\pmb {A}_{N}/N\). By Lemma 3 of Stone (1985) and Lemma A.1 and using Assumption 4 and the fact that \(\chi _{r\nu }(X_{ijr})\chi _{r\nu '}(X_{ijr})=0\) for \(\nu \ne \nu '\), we have that
where \(E_{rM_{rN}\nu }=\sum _{i=1}^{m}\sum _{j=1}^{n}\chi _{r\nu }(X_{ijr})/\sum _{i=1}^{m} \sum _{j=1}^{n}\chi _{rM_{rN}}(X_{ijr})\), \(G_{r}=(g_{rij})_{(p_{r}+1)\times (p_{r}+1)}\) with \(g_{r11}=2\) and \(g_{rij}=\int _{-1}^{1}x_{r}^{i+j-2}dx_{r}-\int _{-1}^{1}x_{r}^{i-1}dx_{r} \int _{-1}^{1}x_{r}^{j-1}dx_{r}/2\) for \(i>1\) or \(j>1\), \(G_{r}^{*}=(g_{rij}^{*})_{p_{r}\times p_{r}}\) with \(g_{rij}^{*}=\int _{-1}^{1}x_{r}^{i+j}dx_{r}-\int _{-1}^{1}x_{r}^{i}dx_{r} \int _{-1}^{1}x_{r}^{j}dx _{r}/2\) and \(\pmb {a}_{r\nu }=(a_{r\nu 0}, a_{r\nu 1},\ldots ,a_{r\nu p_{r}})^{T}\) for \(\nu =1,\ldots ,M_{rn}-1\) and \(\pmb {a}_{rM_{rn}}=(a_{rM_{rn}1},\ldots ,a_{rM_{rn}p_{r}})^{T}\). For fixed \(p_{r}\), it is easy to prove that \(G_{r}\) and \(G_{r}^{*}\) are positive definite. Hence, there exists a positive constant \(C_{1}^{*}\) such that \(\lambda _{min}\ge C_{1}^{*}+o_{p}(1)\) and consequently \(M_{N}\pmb {A}_{N}/N\) is invertible. By arguments similar to those used to prove Lemma A.1 and using the fact that \(E(f_{kr}(U_{r}))=0\) for \(k=1,\ldots ,d_{1}\), we get that
Using Lemma A.2 and assumption 5, we obtain that
Now Lemma A.3 follows from (A.5), (A.20) and (A.21). \(\square \)
Proof of Theorem 3.1
Using (A.13), we obtain that
Similar to the proof of (A.18), we have
Since
then by (A.3), we have
Similarly, \(N^{-1/2}\sum _{i=1}^{m}\sum _{j=1}^{n}(\breve{Z}_{ijk}-V_{ijk})\varepsilon _{ij} =o_{p}(1)\). Under the assumptions of Theorem 3.1, it is easy to prove that
Hence,
Similar to the proof of Lemma A.2, we deduce that
Therefore, Using Lemma A.2 and (A.27), we conclude that
By arguments similar to those used in the proof of Lemma 6 of Tang and Cheng (2009), we can prove that \(N^{-1/2}\sum _{i=1}^{m}\sum _{j=1}^{n}\pmb {V}_{ij}\varepsilon _{ij}\) is asymptotically normal. Therefore, (3.2) follows from (A.4), Lemma (A.3), (A.25) and (A.28). The proof of Theorem 3.1 is completed. \(\square \)
Proof of Theorem 3.2
Let \(\pmb {a}_{0}=(\pmb {a}_{01}^{T},\ldots ,\pmb {a}_{0d}^{T})^{T}\). By arguments similar to those used to prove Lemma A.2, we deduce that
Hence, under the assumptions of Theorem 3.2, by arguments similar to those used to prove Lemma A.2, we obtain that
where \(\pmb {A}_{-r}(\pmb {X}_{ij})=(\pmb {A}_{1}(X_{ij1}),\ldots ,\pmb {A}_{r-1} (X_{ij(r-1)}),\pmb {A}_{r+1}(X_{ij(r+1)}),\ldots ,\pmb {A}_{d}(X_{ijd}))^{T}\). Therefore,
Now by arguments similar to those used in the proof of Theorem 3.1 of Hallin et al. (2004) and using (A.29), we can easily complete the proof of Theorem 3.2. \(\square \)
Rights and permissions
About this article
Cite this article
Qingguo, T., Wenyu, C. Estimation for partially linear additive regression with spatial data. Stat Papers 63, 2041–2063 (2022). https://doi.org/10.1007/s00362-022-01326-8
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00362-022-01326-8