Page 37 - Contributed Paper Session (CPS) - Volume 1

CPS694 Ordak Michal
homogeneity of variance. Almost 62% of respondents do not use statistical
tests to examine the equinumerosity of groups compared, but guesstimate
them instead. Another example is the heterogeneity of variance that may be
related to the occurrence of outliers, zero variance from one of the groups, or
it may be the result of an impact of an independent variable not considered
in the analysis. As many as 49% of respondents do not take account of the
assumption about the homogeneity of variance. As many as 69% people
surveyed do not know the names of non-parametric equivalents of the
statistical tests used, such as the Friedman test.
During the many-year review of thousands of various scientific works, such
as publications, projects, grants, etc., I observed many very significant
statistical errors. They have a significant impact on the scientific value of
published research results. This is the reason for frequent discrepancies in the
results obtained in specific areas. When assessing submitted publications,
reviewers evaluate them primarily in terms of innovation. As an experienced
expert in statistics, I must state that numerous reviewers, even in very high-
ranked journals, accept submitted works despite significant statistical errors.
The problem is the lack of professionals who can review submitted works in
terms of statistics. A significant number of journals included in the ISI Master
Journal List do not ask statistical reviewers for help (Petrovečki, 2009). In order
to illustrate this problem, I assessed the correctness of statistical analyses
carried out in journals included in the ISI Master Journal List. I analysed 30
publications from 50 medical journals (n=1500). Table 1, below, contains basic
statistical errors made by the medical community between 2006 and 2018.
Only 40% of the articles analysed contained correct statistical analyses. The
most common errors I have noticed include: no description or the use of
wrong statistical tests (assumptions of parametric equivalents are not met) and
an incorrect record of results obtained. Type I cumulative error can also be
included here.

Comments on the statistical Example
analysis of results

Results are not recorded according r=0.36 (significance is not provided)
to standards.
Results are not separated by spaces. t(56)=5.69;p<0.05

No statistical tests are used. Only numbers are described and
appropriate statistical tests are not used.
The statistical tests used are either Only the statistical package is described
roughly described or not described at all. without the tests used.

26 | I S I W S C 2 0 1 9