Measurement Error in Dependent Variables in Accounting

Illustrations Using Google Ticker Search and Simulations

Ed deHaan, Alastair Lawrence, Robin Litjens

Research output: Working paperOther research output

Abstract

This paper illustrates how measurement error (“ME”) in dependent variables not only reduces power but, under common conditions in accounting and finance studies, can lead to statistical biases and erroneous inferences. These confounds exist because ME in accounting-based proxies is typically nonadditive, which violates the simple assumptions discussed in most econometrics texts. We demonstrate the effects of nonadditive ME in papers using Google ticker search volume index (“SVI”) as a measure of investor attention. We show that ME in SVI generates both type I and II errors in published studies, and we introduce a new measure of investor-specific ticker search to reduce biases in future research. We also use simulations to show that small amounts of ME in accounting asset values can confound inferences in commonly-used accounting-based proxies such as ROA and Tobin’s Q. Our findings contribute to the literature by improving researchers’ understanding of the effects of ME in common analyses.
Original languageEnglish
PublisherSSRN
Number of pages66
Publication statusPublished - 1 May 2019

Fingerprint

Simulation
Google
Measurement error
Investors
Inference
Finance
Asset value
Econometrics
Type II error
Type I error
Tobin's Q

Keywords

  • dependent variables
  • measurement error
  • bias
  • Google search
  • SVI

Cite this

@techreport{e43caa13bc3b48f795ae74a5060350c7,
title = "Measurement Error in Dependent Variables in Accounting: Illustrations Using Google Ticker Search and Simulations",
abstract = "This paper illustrates how measurement error (“ME”) in dependent variables not only reduces power but, under common conditions in accounting and finance studies, can lead to statistical biases and erroneous inferences. These confounds exist because ME in accounting-based proxies is typically nonadditive, which violates the simple assumptions discussed in most econometrics texts. We demonstrate the effects of nonadditive ME in papers using Google ticker search volume index (“SVI”) as a measure of investor attention. We show that ME in SVI generates both type I and II errors in published studies, and we introduce a new measure of investor-specific ticker search to reduce biases in future research. We also use simulations to show that small amounts of ME in accounting asset values can confound inferences in commonly-used accounting-based proxies such as ROA and Tobin’s Q. Our findings contribute to the literature by improving researchers’ understanding of the effects of ME in common analyses.",
keywords = "dependent variables, measurement error, bias, Google search, SVI",
author = "Ed deHaan and Alastair Lawrence and Robin Litjens",
year = "2019",
month = "5",
day = "1",
language = "English",
publisher = "SSRN",
type = "WorkingPaper",
institution = "SSRN",

}

Measurement Error in Dependent Variables in Accounting : Illustrations Using Google Ticker Search and Simulations. / deHaan, Ed; Lawrence, Alastair; Litjens, Robin.

SSRN, 2019.

Research output: Working paperOther research output

TY - UNPB

T1 - Measurement Error in Dependent Variables in Accounting

T2 - Illustrations Using Google Ticker Search and Simulations

AU - deHaan, Ed

AU - Lawrence, Alastair

AU - Litjens, Robin

PY - 2019/5/1

Y1 - 2019/5/1

N2 - This paper illustrates how measurement error (“ME”) in dependent variables not only reduces power but, under common conditions in accounting and finance studies, can lead to statistical biases and erroneous inferences. These confounds exist because ME in accounting-based proxies is typically nonadditive, which violates the simple assumptions discussed in most econometrics texts. We demonstrate the effects of nonadditive ME in papers using Google ticker search volume index (“SVI”) as a measure of investor attention. We show that ME in SVI generates both type I and II errors in published studies, and we introduce a new measure of investor-specific ticker search to reduce biases in future research. We also use simulations to show that small amounts of ME in accounting asset values can confound inferences in commonly-used accounting-based proxies such as ROA and Tobin’s Q. Our findings contribute to the literature by improving researchers’ understanding of the effects of ME in common analyses.

AB - This paper illustrates how measurement error (“ME”) in dependent variables not only reduces power but, under common conditions in accounting and finance studies, can lead to statistical biases and erroneous inferences. These confounds exist because ME in accounting-based proxies is typically nonadditive, which violates the simple assumptions discussed in most econometrics texts. We demonstrate the effects of nonadditive ME in papers using Google ticker search volume index (“SVI”) as a measure of investor attention. We show that ME in SVI generates both type I and II errors in published studies, and we introduce a new measure of investor-specific ticker search to reduce biases in future research. We also use simulations to show that small amounts of ME in accounting asset values can confound inferences in commonly-used accounting-based proxies such as ROA and Tobin’s Q. Our findings contribute to the literature by improving researchers’ understanding of the effects of ME in common analyses.

KW - dependent variables

KW - measurement error

KW - bias

KW - Google search

KW - SVI

M3 - Working paper

BT - Measurement Error in Dependent Variables in Accounting

PB - SSRN

ER -