data analysis interpretation and causality confounding Flashcards
when should you be concerned about estimates?
with CI - if the CIs show a contradictory estimate
clinically - if the doctor could diagnose something else 95% of time
what do 95% CIs mean?
that 95% of the time the estimate lies within the confidence intervals
what is the p value?
the probability that the p value is at least as big as your assuming the coefficient is actually 0
what does a small p value mean?
zero-assumption is probably wrong and an effect is likely
what does a large p value mean?
a zero assumption is probably right and an effect is unlikely
what is R2?
how good a fit the model is - how well the points align to the line of best fit
what is S?
the deviation of the point
how would you carry out a statistical test?
observe, guess using model, test R2, S, CI and p value and assess using p value, R2 and CI
what is the use of causal inference?
understanding and identifying causal effects help us to understand changing care and improvements
what is association not?
causation
why is causation of infectious disease and cellular processes fairly simple?
bacteria + person = illness
relaxed myometrial cells + prostaglandin E2 = contracted myometrial cells
how are cellular processes used in clinical practice?
relaxed mymometrial cells + prostaglandin E2 = contracted myometrial cells - can be used to induce labour
the deterministic approach to causation is easy. How is it used in life?
it is appealing to toddlers and how they work but is actually inaccurate as association does not necessarily mean causation
what are the multiple causes of pre-term birth?
obesity, smoking, diabetes, alcohol, SGA, country of residence, bacterial vaginosis, iatrogenic etc
why does presentation of data matter?
the way you present and type of graph convinces public in different ways
how do we study how things work?
come up with an estimate of the counterfactual
how can we infer if a change has worked?
can look at different things with different exposures and work out if one variable has more of an effect however everyone is different and there are different environments
what is exchangeability?
when estimating the counterfactual the best way to do this is by finding groups that are comparable through randomisation in a population. This is a biased tool however as you rely on probability to balance out all of the different factors that make people different and therefore need enough numbers to account for the differences and balance them out
what is random sampling error?
it is the random error in out population estimates, that result from chance fluctuations in our sample profile - sample population will never be perfectly representative of a whole population
why wont a bigger sample help to achieve exchangeability without randomisation ?
the exposure without randomisation is assigned by the underlying bio-psycho-social determinants - larger samples want to reduce error but without randomisation this cannot be achieved
what is one of the most important skills in observational research?
learning to identify and address confounding
what is confounding?
it is the distortion of the association between the two variables due to a common shared cause - can generate spurious associations or mask, exaggerate or suppress and association
what are common confounders?
age and sex
what is conditioning?
when we reduce confounding by examining like for like and look specifically at one group - more comparable groups that are exchangeable is conditioning
what is restriction?
when you restrict the sample to a single value of the confounder
what is stratification?
when you calculate category specific effects for different levels of the confounder
what is covariate adjustment?
when you adjust for a covariate in regression
why can conditioning not completely remove confounding?
other confounding variables not considered - unobserved confounding and error in measure - imperfect conditioning and therefore residual confounding
larger samples reduce random error but do not reduce systematic error, what is this?
bias
what is imprecise and inaccurate?
imprecise - lower sample size
low quality - inaccurate
what is the difference between measurement error and measurement bias?
measurement error is error in measurement due to random factors whereas bias is due to non random
when do misclassification errors and bias occur?
when measurement errors and bias are present
what are the types of bias?
selection, confounding, information, experimenter, analytic, inferential - not mutually exclusive
what is selection bias?
occurs due to a systematic difference between those selected and those not for a study - differences between the observed and the unobserved group
what are the types of selection bias?
attrition - loss of participants
participant - who takes par - people having differential preferences or opportunities to take part - willingness varies with all possible bio-psycho-social factors
what is information bias?
when there is systematic error in reporting, measuring or recording information
what types of information bias are there?
response bias - when people answer inaccurately (also acquiescence in this category - inherently more likely to say yes)
recall - people remembering things differently
observer effect - people responding differently when they know they’re being observed
what is experimenter bias?
it can be unconscious or conscious and is due to the behaviours or actions of the experiments
what are the types of experimenter bias?
confirmation - more likely to accept results if fit with what we expect and refute those that dont
systemic - more likely to chase positive associations - p hacking - seek novel results, or publish those that help us
what must be checked in a study?
funding for conflicts of interests to see experimenter bias
what are the two types of error in population research?
random and systematic
what does the sample size have no effect on ?
the degree of systematic error and therefore the accuracy
what is confounding bias?
it is the distortion of the causal association between two variables due to a common shared cause which is known as the confounder
what is a measurement error?
it is an error in measurement due to random factors such as weighing scales varying in climate
what is measurement bias?
it is an error in measurement due to non random factors such as weighing scales broken
what is misclassification error or bias?
it is when measurement error or bias result in misclassification
what is bias?
there are different types and they are not mutually exclusive - a study can be biased in many different ways
what types of bias are there?
inferential, informational, selection, confounder, experimenter and analytic
what are common confounders and when will these occur?
age and sex when there are non standardised rates
