• Statistics
• qualitative
• nominal
• e.g. hair colour
• ordinal
• e.g. ASA
• tests
• 2 groups
• Chi-squared test
• work out the numbers if the null hypothesis was true
• only use numbers in the shoe!
• degrees of freedom = (number of rows-1)-(number of columns -1)
• 95% of Chi squared distribution lies below chi-squared value of 3.842 for 1 degree of freedom
• the greater the Chi-squared value, the more likely the result is to be significant
• Fishers exact test (used for small samples)
• use if expected number in a group is <5
• 3 groups
• don’t need to know!
• quantitative
• continuous/discrete
• distribution
• parametric
• central tendency
• mean
• spread
• variance
• Σ(x-x̄)²/n
• One thing!
• standard deviation
• √Σ(x-x̄)²/n
• 1 S.D. - incorporates 68% of sample, 2 S.D. - incorporate 95% of sample, 3 S.D. - incorporate 99.7% of sample
• Two things!
• standard error of the mean
• SD/√n
• (n = sample size)
• used to make confidence intervals
• 95% CI = 1.96 SEM!
• How sample compares to population
• tests
• 2 groups
• student T-test
• paired (subject + control are the same person)
• unpaired
• 3 groups
• ANOVA
• multiple T-tests
• need to divide P value by number of t-tests as adding in inaccuracy at each stage (BONFERRONI CORRECTION)
• non-parametric
• central tendency
• MEDIAN is best
• mode is highest point
• mean is towards the tail
• ‘mean can hardly be seen!’
• arithmetic mean can be used - but is worst measure
• geometric mean cannot be used
• Skew
• right/positive
• TAIL TO THE RIGHT!
• left/negative
• TAIL TO THE LEFT!
• spread
• range
• interquartile range
• box and whisker diagram
• tests
• 2 groups
• Man-Whitney test
• 2 distinct groups
• Wilcoxon ranked test
• Paired (same) group
• 3 groups
• general
• probability
• P=0.05
• statistically significant
• not necessarily CLINICALLY significant
• randomisation
• removes SELECTION bias
• blinding
• removes OBSERVER bias
• error
• Type 1/false positive/alpha/incorrect rejection of null hypothesis
• type 2/false negative/beta/failure to reject a false null hypothesis
• power = 1-beta
• the chance of picking up a difference if it exists
• normally aim for >80%
• correlation
• coefficient
• -1 —> +1
• +1 = perfect positive correlation
• 0 = no correlation
• -1 = perfect negative correlation
• Pearson
• used for PARAMETRIC data
• Spearman
• used for NON-PARAMETRIC DATA
• regression
• finding best fit of a line
• must pass through MEAN of x AND y data
• NNT
• =1/ARR
• relative risk
• absolute risk reduction
• = control event rate - exposure event rate
• relative risk reduction
• = ARR/control event rate
• risk ratio
• = exposure event rate/control event rate
• Odds
• = number of events/number of non-events
• e.g. rolling a 3 = 1/5
• odds ratio
• = odds in treatment group/odds in control group
• 1/5 / 5/1
• Takes into account the incidence
• sens/spec
• draw the table
• truth along the top!
• remember positive predictive value does the positives only, then its plain sailing...