Week 4 Flashcards
one-tailed hypothesis
a directional hypothesis
specific direction of effect
e.g. greater than .50
two-tailed hypothesis
a non-directional hypothesis
open ended, no set direction
e.g. not equal to .50
alternate hypothesis can be..
directional or non-directional
two-tailed
consider both tails of the probability distribution
one-tailed
look at only one tail of the probability distribution
What should we consider when using a one-tailed test?
easier to find the effect in the expected direction but at the cost of not being able to find the effect in the unexpected direction
if we have strong theoretical grounds
and we can expected a particular effect, then you can consider running a one-tailed test
findings replicated over and over by independent researchers may allow us to make directional predictions
when do we need to use a two-tailed test?
if there isn’t much evidence to suggest data goes in a expected direction
if testing an intervention that is exploratory
if test a intervention that may cause harm
before seeing the data
a decision to use a one or two tailed test must be made and decisions should not be changed
pre-registration
researcher declares their intentions ahead of time, other researchers can then verify
poor scientific bias
- evidence not supporting claims
- misleading media portrayal
- need for open access
- publication bias (hidden trial)
publication bias
a bias towards publishing significant findings so that evidence that finds an effect is reported
studies that don’t find an effect will go unreported, leading to skewed effect
issues
fraud
ways forward
- open access data, everyone has access to all data to be able to test and validate
- confidence by researchers to submit null findings for publications
research hypothesis
a abstract statement of our beliefs about a population parameter
statistical hypotheses
the null hypothesis is a formal statement that the effect we are looking for does not exist in the population
the alternative hypothesis is a formal statements that the effect we are looking for does exist in the population
null hypothesis
the default assumption
assumed to be true until we can falsify it
sampling error
even if we draw two samples from identical populations, they will look different
p value
the statement of probability
how likely would it be to observe results at least as extreme as ours if the null hypothesis were true
alpha level
a criterion
usually .05
if p is less than .05
we reject the null hypothesis
if p is grater than .05
we accept the null hypothesis
type 1 error
false postive
type 2 error
false negative
what does the alpha level control?
the type 1 error rate
one-tailed hypothesis
specifies an expected direction for the effect
two-tailed hypothesis
does not specify a direction of effect
once committed to a one tailed…
you cannot detect a statistically significant effect in the unexpected direction