Abstract
In this paper, we investigate the variable selection for varying coefficient errors-in-variables (EV) models with longitudinal data when some covariates are measured with additive errors. A variable selection method based on bias-corrected penalized quadratic inference function (pQIF) is proposed by combining the basis function approximation to coefficient functions and bias-corrected quadratic inference function (QIF) with shrinkage estimations. The proposed method can handle the measurement errors of covariates and within-subject correlation, estimate and select non-zero nonparametric coefficient functions. With appropriate selection of the tuning parameters, we establish the consistency of the variable selection method and the sparsity properties of the regularized estimators. The finite sample performance of the proposed method is assessed by simulation studies. The utility of the method is further demonstrated via a real data analysis.
Acknowledgments
We thank the editor and reviewers for their helpful comments that significantly improved the manuscript.
Appendix 1. Derivation process of EquationEquation (9)
(9)
(9)
According to the above, we can see that
For simplicity, denote According to
for
so we ca get
where
and
The derivation process of EquationEquation (9)(9)
(9) is finished.
Appendix 2. Proof of theorems
Lemma 1.
Suppose Conditions C2 and C8 hold and , there exists a constant
that satisfies
(22)
(22)
Lemma 1 is the Corollary 6.21 in Schumaker (Citation2007), the proof is omitted here.
Lemma 2.
Assume Conditions C1-C11 hold, and , then we have
where “
” represents the convergence in probability.
where “
” represents the convergence in distribution.
Proof.
We first prove part (i). According to EquationEquation (12)(12)
(12) , we can get the first derivative of
about β as
(23)
(23)
Consider the kth block matrix of as
Now, we prove as
where
Clearly we get
as
According to EquationEquation (10)(10)
(10) , we see that
is the sample covariance matrix of Σu, which implies that
is the mean of some sample covariance matrices and
as
According to the plug-in principle, we get
Under Condition C9, we can get To prove
denote
where
we can get
and
From Conditions C4-C7, is bounded. By the law of large numbers,
So we get
and
where
The proof of the part (i) is completed.
Nextly, we prove part (ii). We first prove Consider the kth block matrix of
as
where
Obviously, we have and
From Conditions C4 to C7, is bounded, and we can get
by the law of large numbers. Similarly, we have
and satisfies.
In addition, According to the Cauchy-Schwarz inequality we have
Therefore,
Under Condition C8 and Lemma 1, we have From the definition of
and by the law of large numbers,
By the definition of
and the central limit theorem, we can get
So we have
and
Next, we prove According to the above conclusions, we have
where
Furthermore, we get
Obviously, according to the above conclusions, we can get
Under conditions C4-C7, following Tian, Xue, and Liu (Citation2014), for any which satisfies
and
Similarly, for any
such that
then
Using the Cauchy-Schwarz inequality, for any
such that
So, we know that
and
satisfy the Lyapunov condition for central limit theorem. In addition, we have
under condition C5, so we get
such that
which implies that
satisfies the Lyapunov condition for the central limit theorem. Thus
According to condition C4, we have So
The proof of Lemma 2 is completed.
Lemma 3.
Suppose that the preceding regularity conditions of C1-C11 hold, then
Proof.
Following Tian, Xue, and Liu (Citation2014), apply Taylor expansion to at
we have
where
is a three-dimensional array of
By Lemma 2, we can see that
Under condition C4, we have
So, we have
Similarly, we get
where
is a four-dimensional array
By the definition of and
so we have
Using Lemma 2, we get
Hence we have So we get
The proof of Lemma 3 is completed.
Proof of Theorem 1.
From Lemma 1, we have Suppose
and
To prove Theorem 1, we just have to show that for any
there exists a large constant C such that
(24)
(24)
When (24) is always true. So we just assume
Without loss of generality, assume
and
we have
Apply Taylor expansion to at
we have
where
lies between β and
According to Lemmas 1 and 2, we can get
and
Therefore we have
Obviously, When C is large enough,
So when C is large enough,
Assume and
When n is large enough, we have
Following the definition of the penalty function, we get
So, for any given there exists a large enough C which satisfies EquationEquation (24)
(24)
(24) , which further implies that there exists
which satisfies
Note that
With the same arguments above, we can get Therefore, invoking
we have
With Lemma 1, we get
Thus, we complete the proof of Theorem 1.
Proof of Theorem 2.
Assume for
and
are non-zero coefficient functions. So we get the corresponding regression parameter space
as
For denote
where 0 is an
vector of zeros. From Lemma 1 and Xue, Qu, and Zhou (Citation2010), we have
and
To prove Theorem 2, it is sufficient to show that, for any
and
is true with probability 1.
where
lies between
and β, t lies between
and
Furthermore, we get
According to Lemmas 2 and 3, we have
Form Conditions C10 and C11,
So for any and
is true with probability tending to 1. This completes the proof of Theorem 2.
Proof of Theorem 3.
Following Wang, Li, and Tsai (Citation2007) and Tian, Xue, and Liu (Citation2014), we create three mutually exclusive sets:
where
and
represent underfitted, correctly fitted or overfitted model
respectively. Then, the theorem can be proved by comparing
and
Here we consider two separate cases.
Case I: When we have
By the law of large numbers and the continuous mapping theorem, we can get
Case II: When
we have
and
So we can get
Both cases hold true in probability by the law of large numbers and the continuous mapping theorem. This completes the proof of the Theorem 3.