Annals of Saudi Medicine: Reply
Reply
================================================================================
Ahmad E. Aboshaiqah, Joel G. Patalagsa on 06/12/2018 09:56:00
Thank you very much Sir for providing your valued feedback to our published
manuscript. We greatly value and appreciate learning from your wisdom and
expertise. We open-heartedly take notice of your points.
1. On why we opted to use non-parametric statistics (Mann-Whitney U test and
Spearman rank correlation coefficient: Classifying data based on the level of
measurement is the basis for the statistical test that we used. When the
dependent variable data is an ordinal level of measurement and assumption of
normality of distribution is violated, nonparametric statistics may be used.1
The reason we used nonparametric statistics (Mann-Whitney U Test and Spearman
rank correlation coefficient) is that we treated the dependent variable
(competence and importance) as an ordinal level of measurement since
participants subjectively rate their competence or importance of a particular
item. This is briefly described in the manuscript under METHODS section from
line 21 to line 25 at the right column on page 289. The aggregated average
therefore is a representation of the subjective ratings of the participants from
the 30 items in the scales. One of the premises in treating our data as ordinal
level of measurement is based on Fieldâ€™s suggestion2 that in situations where
people are asked to rate something subjectively the data should probably be
regarded as ordinal. There are contrasting schools of thoughts on how to
consider the level of measurement of a Likert-type scale data. One school of
thought suggests that Likert-type scale is an ordinal level of measurement but
can be considered an interval level of measurement if there are five response
points in the scale (Johnson & Creech, 1993 as cited by Langdridge &
Hagger-Johnson, 2009, p47).3 On the other hand, Miles and Banyard, 2007, (as
cited by Langdridge & Hagger-Johnson, 2009, p47)3 suggested that Likert-type
scales with at least seven points may be considered an interval level of data.2
This is supported by Tabachnick and Fidell4 who stated that variables are often
treated by statisticians or researchers as if they are continuous when the scale
is thought to be continuous, but apparently the measured scale is ordinal, the
number of scale is large, that is, seven or more and the data meet other
assumptions of the analysis like normality in distribution.4 Based on the
apparent positions of Miles and Banyard (2007) and Tabachnick and Fidell,4
Likert-type scales with less than seven response points are apparently
considered an ordinal level of measurement and represent our second premise for
considering competence and importance scales as ordinal level of measurement
since our items have five response points. In summary, the suggestions of Field2
that subjective ratings may be considered as ordinal level of measurement and
the suggestions of Miles and Barnyard (2007) and Tabachnick and Fidell4 that
Likert-type scales with less than seven response points are considered an
ordinal level of measurement support our premise that competence and importance
scales have ordinal level of measurement. Since our dependent variables
competence and importance are ordinal levels of measurement, we used
nonparametric statistics.
2. We acknowledge that our data may be considered normally distributed based on
a large sam.ple (n=301) as per the central limit theorem.1 This may warrant the
use of parametric statistics. However, we put weight on the level of measurement
if our data as ordinal which called for the use of non parametric statistics.
3. The use of mean in our paper is for the purpose of interpreting the responses
of the participants per item and overall confidence and importance. Since we
used Mann-Whitney U test, the median is used in the comparative data presented
in Table 4.
4. The following table shows the mean range interpretation that we used for each
item and competence and importance as a whole. The purpose of Table 2 and Table
3 is to present how the participants perceived their competence and importance
of the learning objectives included in their training. We believe that plainly
presenting the mean without interpreting it based on the range of responses
would leave users of published studies hanging.
5. The value used in the overall competence and importance is the aggregated
average of all ratings. This is the reason why the median has a decimal point in
Tables 2 (pp.292-293) and Table 3 (pp 293.294).
6. We did not include training hospital and school graduated in the comparison
tests due to the representativeness of participants per category of data. We
could not merge hospitals into two categories since they are all public tertiary
training hospitals and we could not make distinctions. We could not make
credible comparisons of school graduated since the participants in private
schools are relatively few (n=22; 7%) (Table 1, p.291) compared with the number
of participants in public schools. Besides, the figures for perceived confidence
and impotence were presented in the abstract in mean values and the authors
wrote only the P value of significance for the Spearmen correlation and only for
the confidence part of the SECS. I think the authors should clarify in their
methods why they used nonparametric statistics especially when parametric tests
are more robust in this case. Thank you very much for raising your observations.
We believe that the brief presentation at the methods section on page 289 of the
manuscript can be interpreted such that the scales used in the study are an
ordinal level of measurement. We also believe that the presentation of the
statistical test used on page 290 is also brief and specific on what tests were
used and variables included in the analysis. We are sorry for missing the
rs-values in the abstract. This is attributed to the word count limitation. If
the Editor-in-Chief so needed and allow us to have an addendum, we would gladly
add the missing rs-values. We are sorry for missing the r-value for the
importance correlation in the abstract. If the Editor-in-Chief would allow as
have an addendum, we would gladly add the missed result. Kitchen stated in her
paper that the assumptions for nonparametric tests are weaker than those of
parametric and researchers are better to only resort to the first if normality
of data was not met or the underlying distributions are heavily tailed or
extremely skewed and/ or the sample size is small (n
References
1. Weinberg SL, Abramowitz SK. Data analysis for the behavioral sciences using
SPSS. New York: Cambridge University Press. (Reprinted 2007). 2002:257.
2. Field A. Discovering statistics using IBM SPSS Statistics. (5th ed.).
Thousand Oaks, California: SAGE Publications Inc. 2018.
3. Langdridge D, Hagger-Johnson G. Introduction to research methods and data
analysis in psychology. (2nd ed). England: Pearson Education Limited. 2009:47.
4. Tabachnick BG, Fidell LS. Using multivariate statistics. (6th ed.). Upper
Saddle River, New Jersey: Pearson Education Inc. 2013;7.