Primer on statistical reporting and analyses for continuous variables Giuseppe Biondi Zoccai University of Turin, Turin, Italy

Primer on statistical reporting and analyses

for continuous variablesGiuseppe Biondi Zoccai

University of Turin, Turin, Italy [email protected]

SCOPE OF THE PROBLEM

• Continuous variables are very common (e.g. age, left ventricular ejection fraction, hemoglobin concentration, hematocrit, late loss at angiographic analyses)

• Their reporting and analysis strongly depends on the underlying distribution (Gaussian [i.e. normal] versus non-Gaussian)

ALGORITHM

1. Check if variable is well known for having Gaussian distribution

a. If yes and sample size > 30 then proceed to point 2.b. If no or sample size < 30 then proceed to point 3.

2. Report variable as mean±standard deviation and compare it with Gossett-Student t test

ALGORITHM

3. Compare variable distribution with Gaussian distribution using one-sample Kolmogodorov-Smirnov test

a. If p>0.05 then go back to point 2.b. If p<0.05 then go back to point If no or sample size < 30

then proceed to point 4.

4. Report variable as median (1°-3° quartile) and compare it with Mann-Whitney U test (for indepent samples) or Wilcoxon rank-sum test (for related samples)

ALGORITHM

5. If instead than 2 groups, several groups are being compared, other tests should be employed, including ANOVA, MANOVA, or ANCOVA (after transformation) versus Kruskal-Wallis or Friedman non-parametric tests

KOLMOGODOROV-SMIRNOV TEST

KOLMOGODOROV-SMIRNOV TEST

KOLMOGODOROV-SMIRNOV TEST

GOSSETT-STUDENT T TEST

GOSSETT-STUDENT T TEST

GOSSETT-STUDENT T TEST

MANN-WHITNEY U TEST

MANN-WHITNEY U TEST

MANN-WHITNEY U TEST

Thank you for your attention

For any correspondence: [email protected]

For these and further slides on these topics feel free to visit the metcardio.org website:http://www.metcardio.org/slides.html