Page 82 - Academic and Research Book_WEB

contextual variable has to be treated using the stepwise approach (predictor –> contextual variable –>
outcome). This process is called ‘mediation’. It is not necessary that the predictor and outcome have to
be related independently of the contextual variable. Mediation is appropriate to longitudinal studies
and should not be used with cross-section designs.
Instrument quality influences measurements and in turn, influences the inferences drawn.
Measurement error fallacies are associated with aggregation, reliability, and complex constructs.

Summing across individual items would result in compromising the quality of data if: (1) there
is a high proportion of item non-response, and (2) when the scale comprises a mix of positive and
negative constructs. High item non-response would result in underestimation of the total score,
while summing the items. In the case of a mix of positive and negative constructs, it is a common
practice to reverse-code the responses of the negative constructs. However, reverse-coding sums up
with actual codes (positive constructs are untreated and therefore, actual scores of these items are
taken) would incorporate inconsistency in the measurement. Therefore, getting composite scores by
summing individual items should be exercised with caution. There are instances, where the inclusion
of certain variables in a complex structure, results in unintended yet systematic errors. Such variables
are called “nuisance variables”. While sub-scales are used to enhance the reliability, nuisance variables
accidentally get included in the design. These variables need to be identified and their effects are
required to be controlled.

Modern approaches to imputation of the missing data include Full Information Maximum Likelihood
(FIML), the Multiply Imputed (MI), data-based Expectation Maximization (EM) and Markov Chain
Monte Carlo (MCMC) algorithms. The estimates of a statistical model fitted to the data using these
methods are more accurate in comparison with the complete-case analysis or the classical methods of
imputing missing data.
The detailed discussion on the other fallacies, that is, statistical significance fallacies, statistical power
fallacies and factor analysis fallacies can be found in Wang, et al. (2013).

Fallacies in Qualitative Research

Much work has been done in the domain of quantitative research. Qualitative approach to undertake
research has gained acceptance only in the recent past. Thus, relatively less literature is available that
discusses the fallacies in qualitative research.

Fallacies Associated with Descriptive Theorizing

Classifying all measurements into relatively similar and relatively different gives rise to a false
dichotomy. It is recommended that measures should be capable of devising a continuum that gives
information on the relative distance (or similarities) between the cases studied.

There are three issues associated with aggregation are:
1. Lack of clarity about the need to aggregate,

2. Inability to fathom the impact of measurement scales on aggregated scores, and
3. Lack of clarity about the terminology used to refer to the values of the aggregate scales versus
disaggregate scales.
Therefore, the absence of clearly chalked out procedures to undertake aggregation results in conflation
fallacy, when aggregation is performed.

ACADEMIC INTEGRITY AND RESEARCH QUALITY | 71