Lecture 23 Chance 1 Flashcards
Critical thinking
What study is telling you
Start to interrogate
Start to critically evaluate the study and how it was done and what influenced how the study was done will have on the results
External validity
Extent to which you can generalise the results of your study
Judgement call, argue the case
To get external validity
Need internal validity
Internal validity meaning
Do study findings represent the truth
Are they of accurate reflection of the truth or are they misleading in some way
To what extent are they false
Internal validity
Chance
Bias
Confounding
What are we talking about when talking about chance?
Sampling
Sampling
Subset of population (good representation)
whole pop not feasible
what do you use sample for?
estimate of the measure occurrence / association of the population
Estimate population parameter
Parameter
True value in population
Trying to get to know
Estimate
What we get out of our study sample
Study that samples gets an
Estimate of population parameter
Problem of sample
Study samples vary in estimates they provide of the population parameter
If keep randomly selecting people for that study and measuring the same thing
Will get a variety of estimates
Most will be around the parameter
There will be weird outliers
Sample error
Sample varies in the estimates they give us
Some close to the population parameter some will not
Sample error is a form of and occurs by
Form of random error
Occurs by chance
What can be done with this sampling error?
Increase number of people in study (sample size)
Increase sample size
Reduces sample variability (standard deviation etc.)
Increases likelihood of getting a representative sample
Increases precision of parameter estimate
- Increases likelihood that its close to parameter as opposed to further away
More of samples
accurate representation of population parameter
What we mean by chance and sampling error is that our
sample is weird (distorted estimate from population parameter) and estimate isn’t an accurate reflection of the parameter
Sampling
Chance occurs (distorted estimate from population parameter)
Paths might take to deal with a problem of chance and assessing impact on studies
Confidence Intervals
P-values (hypothesis testing)
Confidence Intervals 95%
Contains the true population parameter (value)
5% wouldn’t
give insight into precision
Width of confidence interval shows
How precisely estimating the population parameter
Narrow CI
More precise
Wide CI
Less precise
Increase sample size
More likely get an estimate closer to the parameter
Narrow CI
- As study estimate will most likely be closer to the population parameter
- Decrease width of CI
Use of 95% CI
In assessing clinical importance
- Test for effectiveness of intervention whether it actually helps people
- Often state what is a clinically important difference or clinically meaningful effect
10% reduction risk
= RR 0.9
- Clinically important threshold
CI help us interpret
Study findings from RCT in relation to clinical importance
Any study RCT that produces RR lower than 0.9 closer to 0
Meet threshold
Produce clinically important effect
Any study RCT that produces RR above 0.9
Does Not meet clinical importance threshold
Tight CI below threshold
Study is clinically important
Estimate below threshold
CI below threshold
Population parameter below threshold
Wider confidence interval
Lowest end (more effective if PP there), or less effective of what threshold is and be a slight risk factor Possibly clinically important Does Not contribute a great deal to knowledge
Tight CI above clinically important threshold (CI spans below threshold across to just below null)
Consistent with reduction of risk below null
Crosses clinically important threshold
Possibly clinically important
Tight CI above clinically important threshold
Precisely estimates
Not clinically important
CI doesn’t meet threshold
Tight CI above important threshold
CI inconsistent with population parameter
When sampling
Generate estimate of population parameter
Parameter may truly never know
Estimating population parameter for sampling
Introduces random error and sampling error and chance
Just by chance
Might get samples that produce an estimate really weird
Isn’t a great reflection of population parameter
Can’t eliminate only reduce by increasing sample size
A way to assess how precisely and reduce the influence of chance
Use 95% confidence intervals
Tells value between that we are 95% confident that the true value population parameter lies
Precision
How precisely estimating population parameter
How close likely to be to it with estimate
If wide CI
Study has a wide range of interpretations
Find point estimate a measure associations something is a risk factor but CI being consistent with population parameter of the exposure being a protective factor
Precision is important for
What we take away from our measures of association
Larger sample size
Better for CI and precision
Narrow CI
Great
Need larger sample size
chance consists of
validity (ext and int)
CI
sampling error
External validity
Generalisability - The extent to which the findings of the study be applied.
Judgement call depending on what is being studied and who it is being applied to
Internal validity
The extent to which the findings of the study are free of chance, bias and confounding
Are there other explanations for the study findings, apart from them being right?
pop parameter
True value of measure
we’re interested in
sample estimate
Study’s estimate of the parameter
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
If you repeatedly sampled randomly from the same population,
get a sample with a similar composition to the population you sampled from
some samples would be different by chance
sampling error, and is a form of random error known as chance
Narrow CI
more precise
RR = 2.0 (95% CI: 1.8 – 2.2)
Wide CI
less precise
RR = 2.0 (95% CI: 0.8 – 3.2)
Increasing the sample size makes the
CI narrower