Enyi Patrick Enyi
Babcock University
e.p.enyi@gmail.com
Department of Accounting, School of Management Sciences
Accounting and finance-based researchers often face the problem of using
multiple surrogates to capture the true essence of a dependent variable
in order to study its predictive relationship with explanatory
variables. The major pitfall with this, is the inability to directly
connect the study results with the major objective of the research
forming the topic. This study explores the various methods that can be
adopted to unify the engaging properties of the multiple parameters used
in identifying a dependent variable. Using logistic regression, the
study converted the a priori expectations of 30 Ph.D research
theses in finance and accounting with four dependent surrogates into a
probabilistic log values and compared them with the individual surrogate
performance on the one hand and the surrogates geometric mean on the
other hand. While the geometric mean showed no significant difference
between itself and the probabilistic expectation (β = .278,t(30) = .695, R2 = .077,p > .10), the individual surrogates results revealed
both combined and individual significant differences with the thesesa priori expectations (Adj. R2 = .0291,F (4, 25) = 22.598, p < .05).
This paper recommends unifying multivariate dependent variables with
geometric means for successful financial performance relational studies.
Keywords : performance, dependent variable, surrogate, proxy,
logistic regression
Researchers and scholars in the social and management sciences are often
faced with the problem of identifying and operationalizing the main
ingredients forming the dependent or outcome variable in performance
assessment research. This is particularly a big issue when such a study
is extended in scope or necessitated by the requirements to fulfil the
conditions for the award of a degree. For those studying behavioral
traits and similar psychological phenomena that has no accounting or
financial data foundation, this will not present a problem as there are
several methods already developed to tackle the issue. Some of these
methods include the use of general linear model (GLM) multivariate and
repeated measures procedures using computer statistical software, while
others may include the use of canonical correlations and specially
constructed regression equations (Helwig, 2017; NCSS, 1989; Steiger,
n.d.). The use of means as a converging point was also advocated by
other researchers especially when multiple dependent variables represent
different measures of the same construct. When such differing variables
are measured on the same scale, researchers have the option of combining
them into a single measure of that construct. The use of binary response
model (BRM) was also in the picture, but as posited by Cameron (2015),
the description of the use of means looks more practical and less
ambiguous to comprehend (Cameron, 2015; Gruszczy, 2009; Price,
Jhangiani, & Chiang, 2015). However, the imprecise and unsure nature of
the various methods in current use makes most research findings and
conclusions meaningless and unrelated to the major objective of the
research. This is very worrisome especially to graduate level research
students and their advisors and has proven to be of particular negative
implications in the fields of accounting and finance in management
sciences where precision in conclusions is of utmost importance in
guiding investors and decision makers correctly.