Page 449 - Contributed Paper Session (CPS) - Volume 2
P. 449
CPS1922 Imran K. et al.
software without the supervision of statistical collaborator are discussed
below.
Case 1) Assigning value zero in the blank cells makes the sample sizes as equal
which in turn increases the degrees of freedom, balances the data, and
decreases standard error. The increase in degrees of freedom has an effect on
the probability value (p-value) and can lead to Type-II error (rejecting false H0).
Treating sample sizes equal will result in inappropriate use of the statistical
tool. In some cases assigning value zero to a blank cell can act as an outlier
and can have a drastic effect on the entire results which can lead to wrong
research answer to the right research question.
Case 2) Since the type of statistical tool to be used is governed by the type of
dependent and independent variable many times it has been observed that
researchers use incorrect statistical tools and thus spoiling the data of good
quality and publish their research in a low ranked journal instead. For example,
while studying Dead/Alive as the dependent variable which is binary along
with other independent variables the researcher apply t-test rather than a
logistic regression.
Case 3) Factorial experiments: In single factor experiments, it has been
observed that in post hoc tests critical difference is used in unplanned pairwise
comparison which increases the probability of significance thus leads to invalid
conclusions. In Factorial experiments, it has been observed that in presence of
significant interaction the main effects are discussed which leads to wrong
information.
Case 4) We know that to calculate odds ratio is easy but its inference is very
tricky and quite often leads to irrelevant inferences/conclusions/
interpretations. By interchanging the rows or columns or both, results in a
change in interpretation of odds ratio.
Case 5) Since for calculating the correlation coefficient we should have a
paired observation on the same unit, sometimes the researcher records the
paired observations but doesn’t place them in pairs while entering the data.
Thus, calculating an invalid correlation.
Case 6) It has been found in the publications that author presents very low R2,
high standard error hence significant regression coefficients. The reason being
that while applying regression models the researcher hardly has any idea of
autocorrelation or multicollinearity or homoscedasticity and makes inferences
about coefficients which are much inflated than actual. Since the innocent
Software has no knowledge about all these things hence can make researcher
SIGNIFICANTLY happy or unhappy as the case may be.
438 | I S I W S C 2 0 1 9