A non-parametric test for partial monotonicity in multiple regression

M. van Beek, H.A.M. Daniëls

Research output: Contribution to journalArticleScientificpeer-review

Abstract

Partial positive (negative) monotonicity in a dataset is the property that an increase in an independent variable, ceteris paribus, generates an increase (decrease) in the dependent variable. A test for partial monotonicity in datasets could (1) increase model performance if monotonicity may be assumed, (2) validate the practical relevance of policy and legal requirements, and (3) guard against falsely assuming monotonicity both in theory and applications. To our knowledge, there is no test for this phenomenon available yet. In this article, we propose a novel non-parametric test, which does not require resampling or simulation. It is formally proven that the test is asymptotically conservative, and that its power converges to one. A brief simulation study shows the characteristics of the test. Finally, in order to show its practical applicability, we apply the test to a dataset and interpret its results.
Original languageEnglish
Pages (from-to)87-100
JournalComputational Economics
Volume44
Issue number1
Early online date21 May 2013
DOIs
Publication statusPublished - Jun 2014

Fingerprint

Monotonicity
Nonparametric test
Multiple regression
Simulation
Ceteris paribus
Simulation study
Practical relevance
Resampling

Cite this

@article{c1baaa6e49a74ae28a0d97f809e9a80e,
title = "A non-parametric test for partial monotonicity in multiple regression",
abstract = "Partial positive (negative) monotonicity in a dataset is the property that an increase in an independent variable, ceteris paribus, generates an increase (decrease) in the dependent variable. A test for partial monotonicity in datasets could (1) increase model performance if monotonicity may be assumed, (2) validate the practical relevance of policy and legal requirements, and (3) guard against falsely assuming monotonicity both in theory and applications. To our knowledge, there is no test for this phenomenon available yet. In this article, we propose a novel non-parametric test, which does not require resampling or simulation. It is formally proven that the test is asymptotically conservative, and that its power converges to one. A brief simulation study shows the characteristics of the test. Finally, in order to show its practical applicability, we apply the test to a dataset and interpret its results.",
author = "{van Beek}, M. and H.A.M. Dani{\"e}ls",
year = "2014",
month = "6",
doi = "10.1007{\%}2fs10614-013-9386-7#",
language = "English",
volume = "44",
pages = "87--100",
journal = "Computational Economics",
issn = "0927-7099",
publisher = "Springer Netherlands",
number = "1",

}

A non-parametric test for partial monotonicity in multiple regression. / van Beek, M.; Daniëls, H.A.M.

In: Computational Economics, Vol. 44, No. 1, 06.2014, p. 87-100.

Research output: Contribution to journalArticleScientificpeer-review

TY - JOUR

T1 - A non-parametric test for partial monotonicity in multiple regression

AU - van Beek, M.

AU - Daniëls, H.A.M.

PY - 2014/6

Y1 - 2014/6

N2 - Partial positive (negative) monotonicity in a dataset is the property that an increase in an independent variable, ceteris paribus, generates an increase (decrease) in the dependent variable. A test for partial monotonicity in datasets could (1) increase model performance if monotonicity may be assumed, (2) validate the practical relevance of policy and legal requirements, and (3) guard against falsely assuming monotonicity both in theory and applications. To our knowledge, there is no test for this phenomenon available yet. In this article, we propose a novel non-parametric test, which does not require resampling or simulation. It is formally proven that the test is asymptotically conservative, and that its power converges to one. A brief simulation study shows the characteristics of the test. Finally, in order to show its practical applicability, we apply the test to a dataset and interpret its results.

AB - Partial positive (negative) monotonicity in a dataset is the property that an increase in an independent variable, ceteris paribus, generates an increase (decrease) in the dependent variable. A test for partial monotonicity in datasets could (1) increase model performance if monotonicity may be assumed, (2) validate the practical relevance of policy and legal requirements, and (3) guard against falsely assuming monotonicity both in theory and applications. To our knowledge, there is no test for this phenomenon available yet. In this article, we propose a novel non-parametric test, which does not require resampling or simulation. It is formally proven that the test is asymptotically conservative, and that its power converges to one. A brief simulation study shows the characteristics of the test. Finally, in order to show its practical applicability, we apply the test to a dataset and interpret its results.

U2 - 10.1007%2fs10614-013-9386-7#

DO - 10.1007%2fs10614-013-9386-7#

M3 - Article

VL - 44

SP - 87

EP - 100

JO - Computational Economics

JF - Computational Economics

SN - 0927-7099

IS - 1

ER -