Statistical analysis guidelines - advances.umw.edu.pl

1. Authors must avoid presenng their results based only on hypothesis tesng and p-value reporng. Effect size measures are also required, e.g. differences between two means (if a sample mean was used as an esmator of the locaon parameter), odds raos (OR), correlaon coefficients, relave risk (RR), coefficient of determinaon in regression (R 2 ) analysis, etc. For most of the presented effect sizes their confidence intervals (CI) will be required to be reported on standard significance coefficient, e.g. α=0.05, so CI95% is appropriate in most cases (0.01 ≤ α ≤ 0.1). 2. The results should be stascally described by basic descripon stascs, e.g. min(x), max, median, mean, first and third quarle, standard deviaon, median absolute deviaon, number of observaon (n), etc. 3. Numerical results must be reported in open form, i.e. in the tables. 4. Parametric methods are preferred over nonparametric rang tests. If data doesn’t meet the criteria for preferred parametric method, e.g. observaons are not symmetrically distributed, heavy one-tailed distributed, data should be transformed first (e.g. with Box-Cox method) and informaon about the type of the transformaon must be included in the Stascal analysis secon. 5. In case of data analyzed with ANOVA, predefined (a priori) comparisons (contrasts) are preferred over unplanned (a posteriori) comparisons. Whenever the k condions (level of a factor) can be arranged from the smallest to the largest level, polynomial orthogonal contrasts should be used and tested on one de- gree of freedom in nominator, instead of mul comparisons aﬅer omnibus test of ANOVA one obtained as significant. Full tables of ANOVA must be included either in the main text or in supplementary materials but supplementary files are preferred. 6. Authors should avoid presenng comparisons of different groups by plong a bar-plots with means and standard deviaons (as well as standard error of the mean, SEM) as a whiskers. Such informaon must be presented in the tables. Instead of this, when a sample size is sufficient, box-and-whiskers plots can be used to graphically describe and compare a data. Otherwise points can be presented individually on the plot with average level (mean, median, etc.). 7. P-values must be presented as first four decimal places, e.g. p = 0.47965415 expressed as p = 0.4796. In case of very low values, e.g. p = 0.0000855, p-values must be expressed as p < 0.0001. Authors should avoid to perform to many and unnecessary stascal comparisons with hypothesis tesng. Instead, effect sizes should be presented with confidence intervals. Not significant results must also be reported to avoid bias in publicaons and future meta analysis. 8. If you use stascs on degrees of freedom (df) (t-test, ANOVA, chi-square test etc.), please report not only p-values but also df number. 9. Whenever (in text, figures, or tables) P-value is given, it should be accompanied by the name of the test (e.g., Mann-Whitney test, or General Linear Model) and the test value (U, t, F, etc.). Statistical analysis guidelines

Upload
others
Category

Documents
view
10
download
0

Embed Size (px):

Transcript of Statistical analysis guidelines - advances.umw.edu.pl

Page 1: Statistical analysis guidelines - advances.umw.edu.pl

1. Authors must avoid presenting their results based only on hypothesis testing and p-value reporting. Effect size measures are also required, e.g. differences between two means (if a sample mean was used as an estimator of the location parameter), odds ratios (OR), correlation coefficients, relative risk (RR), coefficient of determination in regression (R2) analysis, etc. For most of the presented effect sizes their confidence intervals (CI) will be required to be reported on standard significance coefficient, e.g. α=0.05, so CI95% is appropriate in most cases (0.01 ≤ α ≤ 0.1).

2. The results should be statistically described by basic description statistics, e.g. min(x), max, median, mean, first and third quartile, standard deviation, median absolute deviation, number of observation (n), etc.

3. Numerical results must be reported in open form, i.e. in the tables.

4. Parametric methods are preferred over nonparametric rang tests. If data doesn’t meet the criteria for preferred parametric method, e.g. observations are not symmetrically distributed, heavy one-tailed dis-tributed, data should be transformed first (e.g. with Box-Cox method) and information about the type of the transformation must be included in the Statistical analysis section.

5. In case of data analyzed with ANOVA, predefined (a priori) comparisons (contrasts) are preferred over unplanned (a posteriori) comparisons. Whenever the k conditions (level of a factor) can be arranged from the smallest to the largest level, polynomial orthogonal contrasts should be used and tested on one de-gree of freedom in nominator, instead of multi comparisons after omnibus test of ANOVA one obtained as significant. Full tables of ANOVA must be included either in the main text or in supplementary materials but supplementary files are preferred.

6. Authors should avoid presenting comparisons of different groups by plotting a bar-plots with means and standard deviations (as well as standard error of the mean, SEM) as a whiskers. Such information must be presented in the tables. Instead of this, when a sample size is sufficient, box-and-whiskers plots can be used to graphically describe and compare a data. Otherwise points can be presented individually on the plot with average level (mean, median, etc.).

7. P-values must be presented as first four decimal places, e.g. p = 0.47965415 expressed as p = 0.4796. In case of very low values, e.g. p = 0.0000855, p-values must be expressed as p < 0.0001. Authors should avoid to perform to many and unnecessary statistical comparisons with hypothesis testing. Instead, effect sizes should be presented with confidence intervals. Not significant results must also be reported to avoid bias in publications and future meta analysis.

8. If you use statistics on degrees of freedom (df) (t-test, ANOVA, chi-square test etc.), please report not only p-values but also df number.

9. Whenever (in text, figures, or tables) P-value is given, it should be accompanied by the name of the test (e.g., Mann-Whitney test, or General Linear Model) and the test value (U, t, F, etc.).

Statistical analysis guidelines