||What does this mean?
||What kind of study
is most likely to be effected?
|An event that occurs during
a study that can effect the responses of the participants.
It could be something the participants due like start
exercising all on their own or it could be an event such
as an article in the newspaper or national publicity.
Longitudinal. A study with
|The participants get older,
wiser, more depressed, more hungry, more tired. They change
all on their own!
||Longitudinal., studies where
the participants are more likely to change such as adolescents,
infants or people who are quite ill.
|Sometimes the effect being measured changes
because of the number of times the participant is tested.
The test itself may add to the person's knowledge or change
their attitudes, hence the test becomes part of the intervention!
||Studies with repeated measures or a pre-test-post
|This means changes in the instrument during
the study. It can also mean the data collectors get better
or worse at what they are doing. To me this always seems
like a reliability issue, but it is always listed under
Physiologic instruments or those in
which researchers are collecting the data in person.
||In general, participants who score really
high or really low on a test or a questionnaire will have
a more moderate score the next time they are tested. This
is called regressing towards the mean.
||Pretest-posttest designs. A study with
a small sample size will have more of an effect since
with less participants, one person's scores has a greater
effect on the average score.
||Pre-existing differences between the participants
selected for a study or those who volunteer for a study
and those not in the study or differences between or among
the study groups.
||Quasi-experimental studies and convenience
samples. Doing statistical comparisons between groups
and help address this threat.
|Mortality or attrition
||No, this doesn't mean the sicker people
die! Participants drop out of s study or are lost to follow-up.
||Longitudinal. studies. It helps to have
several contact phone numbers or addresses for each participant.
It is important to analyze the existing data to determine
if there are differences between those who dropped out
and those who continued. If more than 10% of participants
are lost to follow-up this seriously effects generalizability.
||One or more of these threats can combine
and have have a compounding effect.
||The three most likely interactions are
history, maturation and instrumentation.
|Nonadherence/failure to complete protocol
||This in essence could water down the effect of the intervention
Studies with an intervention. Especially at risk would be longitudinal studies and studies with a complex intervention.
|Diffusion of treatment
||"Contamination" of the no-treatment
group by the treatment group. Somehow the no-treatment
finds out what the treatment is and start doing it! It
may also mean that the researchers change one of the intervention
groups or control group to contain treatments not originally
in their intervention.
||Studies with an intervention when the intervention
groups and control groups are being studied at the same
time in the same location (a clinic, small town). This
means there is essentially no control group. I would consider
this a fatal flaw! The trade off is the effects of history
and selection bias.
|Compensatory rivalry of the no-treatment
||Participants know they are in the no-treatment
group and so attempt to improve on their own. Good for
them, but bad for the study!
||Intervention studies where the participants
are aware which group they are in. It can also occur if
the health care providers know the participants are in
the no-treatment group so they give better care. Blinded
and double blinded studies help.
|Resentful-demoralization of the no-treatment
||The opposite of above. The control group
gives up because they aren't getting the treatment.
||The reverse of the above. above.
Note: One sited web site
lists these as internal validity threats and they are
usually included under threats to external validity.
|Experimenter bias, Placebo
Effect, Hawthorne Effect.