ELSEVIER Statistics & Probability Letters 34 (1997) 67-73
STATIITICS t
PROBAIIILITY
LETll[R$
Consistent estimation for the non-normal ultrastructural model
Anil K. Srivastava a'*, Shalabh b
aDepartment of Statistics, University of Lucknow, 15 Shastri Nagar, Lucknow 226004, India
b Department of Statistics, University of Jammu, Jammu, India
Abstract
The present article considers the linear ultrastructural model which encompasses the two popular forms of measure-
ment error models, viz., the functional and structural models. The measurement error variance associated with the
explanatory variable is assumed to be known which leads to consistent estimators of parameters. The efficiency
properties of the thus obtained estimators of slope parameter, intercept term and the measurement error variance of
study variable are derived under non-normal error distributions and the effects of departures from symmetry and
peakedness of the error distributions are studied.
Keywords: Immaculate estimator; Measurement errors; Ultrastructural model
1. Introduction
Dolby (1976) introduced the ultrastructural model which includes as special cases the functional and
structural measurement error models. Assuming normal measurement errors, he derived the maximum
likelihood estimates of the parameters. These parameters were found to be unidentifiable due to the lack of
sufficient information. However, he obtained the maximum likelihood estimates of the parameters assuming
the two variance ratios to be known - one is the variance of measurement errors associated with observed
study variable divided by the variance of observed explanatory variable while the other is variance of true
explanatory variable divided by measurement error variance of observed explanatory variable. Cheng and
Van Ness (1991) incorporated additional information in the form of known measurement error variance
associated with explanatory variable and derived asymptotic properties of the maximum likelihood
estimators under the assumption of normal errors. These estimators were found to be consistent. However,
the assumption of normal errors restricts the utility of the asymptotic results in many applications. This
article provides asymptotic results for such estimators when the measurement errors are not necessarily
normal. Such an investigation can shed light on the robustness of the results under non-normal measurement
errors.
* Corresponding author.
0167-7152/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved
PII S0 1 67-7 1 52(96)001 67-8