PAS - Sample Estimates & Confidence Intervals Flashcards
what is a confidence interval?
alternative to p-values
provides a range of values where we can be X% confident the population parameter lies
potential biases that can arise from estimates based on population samples? (3)
selection bias = sample isn’t representative of population
measurement bias = errors in measuring outcomes arise - e.g. from using an inaccurate instrument
publication bias = when positive results are more likely to be published than negative or null results, skews evidence base
how does a smaller sample size affect the confidence interval?
smaller sample size = wider confidence interval = greater uncertainty with estimate
how does a larger sample size affect the confidence interval?
larger sample size= narrower CL = more precise estimates
describe CL interpretation around absolute differences
CL around an absolute difference directly shows the range of possible values for the difference in outcomes
e.g. 95% CL of 2-8% for an absolute risk reduction of 5% = 95% confident the true risk reduction is between 2-8%
describe CL & risk ratios
CL for a (risk) ratio shows the range of possible values for the ratio of outcomes between groups
e.g. relative risk ratio for developing a disease with a new treatment is 0.75 with a 95% CL between 0.6-0.9 = 95% confident that the true relative risk lies between 0.6-0.9
- treatment reduces risk by 25%
calculating & interpreting risk ratio
risk of outcome with treatment divided by risk of outcome with placebo = produces a unitless ratio
RR = 1 - no outcome
RR < 1 - reduction in risk = ratio can be represented as a negative percentage
e.g. 0.91 RR
= -9% proportionate reduction in risk
= -0.09% absolute reduction in risk (per 100 people)
key concepts in the study design of clinical trials
randomisation - lessens selection bias, ensures comparable groups
control group
blinding
placebo-controlled group = helps measure true effect of the intervention
ethical considerations
point estimates - interpretation?
provides a single best estimate of the effect of treatment
confidence intervals - interpretation?
indicate the range of values within which the true effect likely lies
probability/ p-values - interpretation?
assess the evidence against the null hypothesis - lesser than the alpha threshold suggests a statistically significant difference
compare 95% reference range vs 95% confidence interval?
(95%) reference range = measure of the spread of the continuous numerical data only
- mean +/- 2 standard deviations
(95%) confidence interval = measure of the precision of a sample estimate (95% chance that the interval contains the true population value for a certain parameter)
- mean +/- 2 standard errors
standard deviation vs standard error?
SD = variability within a single sample, calculated from sample data
SE = variability across multiple samples of a population, estimated value