Statistics Flashcards
What indicates if a test is reliable?
If it is repeated again and again and gets the same result
What does it mean if a test is valid?
If it meets the requirements of the scientific study
e.g. randomisation, blinding
What type of data is nominal?
categorical, with no ranking
e.g. gender, blood group
What type of data is ordinal?
categorical with ranking
e.g. preference scale
what type of data is discrete?
numerical data which is in whole numbers
e.g. number of people
What is the variance of a sample? How is it calculated?
A measure of dispersion of sample
Average of squared differences from the mean
What is the relationship between variance and SD?
SD = square root of variance
What assumptions must be made with parametric data?
Continuous data
Population data is normally distributed
Sample and source population have same SD/variance e.g. spread
When are non-parametric tests more appropriate than parametric?
When data is not continuous (e.g. ordinal, nominal)
When distribution of population is not known
Small sample size -> less affected by outliers, uses median rather than mean. RANKS data
What is the central limit theorem?
In a skewed population, sample becomes more “normal” in shape as n increases
What is the parametric test used to compare 2 independant groups?
What is the non-parametric equivalent?
T-test
Wilcoxon rank sum test
What is the parametric test used to compare paired observations?
What is the non-parametric equivalent?
T-test for paired
Wilcoxon signed rank test
What is the parametric test used to compare several groups?
What is the non-parametric equivalent?
ANOVA
Kruskal Wallis test
What is the parametric test used to find linear relationship between 2 variables e.g. BP and sleep duration?
What is the non-parametric equivalent?
Pearson’s correlation
Spearman’s Rank correlation
What is the non-parametric test used to test association between 2 qualitative variables e.g. gender, smoking status?
Which one for sample size
>50?
<50?
Chi squared
> 50 -> chi squared
<50 -> Fisher’s exact
What is the difference between t-statistic and z-statistic?
T = sample with unknown SD
Z = known SD
What is a Type 1 error?
How is it minimised?
False positive
i.e. wrongly rejecting null hypothesis
Avoid by setting p-value low enough
What is a Type 2 error?
How is it minimised?
False negative
i.e. wrongly accepting null hypothesis
Avoid by having enough POWER
What is “power” of a study?
i.e. what does 80% power in a study mean?
The ability of a study to demonstrate a statistically significant association
80% power = study has 80% chance of ending with a p-value <0.05
What is “power” of a study?
i.e. what does 80% power in a study mean?
The ability of a study to demonstrate a statistically significant association
80% power = study has 80% chance of ending with a p-value <0.05
What would changing your p-value cut-off from 0.05 to 0.01 do to the power of a study?
Decrease it
Smaller p-value requires more power
How can power be increased in a study design?
Increase sample size
Increase p-value
Increase effect size
What does a 95% confidence interval indicate?
we are 95% confident that true population mean occurs within this interval
What error is more likely with post-hoc analyses?
Type 1 error
What is the familywise error rate?
What is the Bonferroni correction?
Cumulative Type 1 error
e.g. if testing 3 hypotheses, need to calculate 0.95 x 0.95 x 0.95
Bonferroni correction compensates for this by fixing p-value across tests
WHich of these measures is not affected by prevalence of an event?
ARR
NNT
RR
RR
What is the difference between odds of an event and probability of an event?
Odds = no. of events / no. of NON-events
Probability = no. of events / TOTAL events
What is the log rank test used for?
To compare 2 survival curves
Which study design is best for rare disease?
Which study design is best for rare exposures?
Rare disease - case-control
Rare exposure - cohort
What is the rationale for randomisation in RCTs?
To reduce confounding
Reduces selection bias
What is the rationale for blinding?
To reduce information (observe) bias
What kind of bias does non-random sampling lead to?
Selection bias
What kind of bias can occur if investigators know which intervention the next participant will receive
Allocation / channelling bias
What is ascertainment bias?
When data is collected in a way such that some members of population are less likely to be included
e.g. requiring a doctor’s visit to get Alzheimer’s diagnosis
What is information bias?
Systematic difference in way information is collected / recorded
Which types of studies are liable to recall bias?
Where self-reporting occurs e.g. case control
What is the Hawthorne effect?
Participants report improvement if they know they are on the drug/being observed
what is the Halo effect?
error in reasoning where impression is formed due to a single trait
What are trials of nutrition or exercise liable to (with bias)?
Performance bias
may inflate estimated effect of intervention, with subjective outcomes
What is length time bias?
Diseases with long duration are more likely to be included in screening, and are less aggressive
e.g. slower-growing cancers are picked up on screening and tend towards better prognosis
What is lead time bias?
Occurs when a disease is diagnosed earlier, but it actually has no impact on outcome of disease. However it appears like there is prolonged survival
What is the aim of intention to treat analysis?
To minimise selection bias
Under-estimates effect
What is another name for True positive rate?
Sensitivity
What is another name for True negative rate?
Specificity
Which of these are affected by prevalence of disease?
Sens
Spec
NPV
PPV
NPV -> increasing prevalence decreases NPV
PPV -> increasing prevalence increases PPV
What is the purpose of Phase 1 trials?
Focus on pharmacology
- in small number of healthy subjects
- dose-finding, dosing schedule
What is the purpose of Phase 2 trials?
Focus on safety
- usually in subjects with disease, but small number
What is the purpose of Phase 3 trials?
Focus on efficacy
- large numbers of patients with disease
What is the purpose of Phase 4 trials?
Focus on long-term effects
- post-marketing