Chance - I Flashcards
What is internal validity?
Asking ourselves if our study estimate an accurate estimate of the actual value in the source population?
Use study population to estimate things about the source population.
- Source population: actual prevalence/incidence/RR/RD/OR
- Study population: estimated prevalence/incidence/RR/RD/OR
Need to consider:
- bias, chance, confounding
The crux of internal validity is ‘are there any other explanations for the study findings, apart from them being right?
What is external validity?
The extent to which the study findings are applicable to a broader or different population
Also known as generalisability. The ability to generalise the study findings to other populations.
- judgement depending on what is being studied and who it is being applied to
Describe estimating population parameters
- we use an estimate for a parameter of the population so unless we measure the entire population, we will never truly know what the true population value is
describe how study samples vary and what can be done about this
if you repeatedly samples randomly from the same source population, most of the time you would get a sample with a similar composition to the population you sampled from
- but some of the samples would be quite different just by chance
- this is called sampling error and is a form of random error (commonly called chance)
to minimise this we can make the study sample size larger (less spread, much closer to the true value) this:
- reduces sample variability (std. dev. ect.)
- increases likelihood of getting a representative sample
- increases precision of parameter estimate
BUT, we still don’t know if out study is one of the outliers that are not as close to the true value…
What do we mean by chance and sampling error and how do we measure the influence of sampling error?
What we mean by chance and sampling error is that our sample is weird just by chance and therefore our estimate isn’t an accurate reflection of the parameter
- can’t eliminate sampling error by can reduce with larger sample sizes
Measure the influence of sampling error by:
- confidence intervals
- p-values (hypothesis testing)
What is that statistical definition of a 95% confidence interval?
If you repeated a study 100 times with a random sample each time, and got 100 estimates and 100 confidence intervals, in 95 of the 100 studies that parameter would lie within that study’s 95% confidence interval.
In 5 studies, the parameter (true value) would not lie within the study’s 95% confidence interval.
- A 95% confidence interval represents the range of values within which the parameter will lie 95% if the time if we continue to repeat the same study with new samples
Less precise but fine interpretation: ‘We are 95% confident that the true population value lies between the limits of the confidence interval’
- we know the study population confidence interval, so we can infer that the true value for the source population will be somewhere in the confidence interval
How do we interpret a 95% confidence interval and how can it be applied to any numerical measure?
We are ‘confidence interval level’ confident that the parameter lies between ‘lower bound’ and ‘upper bound’
Can be applied to any numerical measure:
eg. we are 95% confident that the true odds ratio lies between 1.1 and 1.4
eg. we are 95% confident the the true prevalence lies between 68% and 84%
how do you use confidence intervals to increase the precision of your perimeter estimate?
the estimate is important, but to understand its implications for practice you need to know how precise it is.
- if you were just looking at the measure of assosciation and it was the same between studies you would think it was definitely correct, however is you add in the confidence intervals and one is massive (it will be less reliable) than one that is smaller.
- we use confidence intervals to increase the precision of the parameter estimate by looking at clinical importance
- increasing the sample size makes the CI smaller (shows more precision)
what is clinical importance?
- RCTS often state what they consider to be a clinically important result
- confidence intervals can help us decide whether the study finding are clinically important or not
- Find the measure of association
- Then plot a graph with the confidence interval.
- if the confidence interval is completely below the measurement then it is clinically important
- if the confidence interval is partly above and partly below the measurement then it is possibly clinically important
- if the confidence interval is completely above the measurement then it is not clinically important
parameter vs. estimate
parameter: the true value of the measure in the population that the study is trying to discover
estimate: the measure found in the study sample. sometimes referred to as the point estimate
