PSYC3042 Final: PSYC3042: Terms S
Only: S
Sequencing
Confound
Exposure to the IV or DV (or associated procedures) causes changes in
responding that provide an alternative explanation for the results (e.g.,
persistent effects of drug dose, practice, or fatigue)
Caused by IV level sequence, not DV measurement
Would change with changes of the sequencing of events (blocked conditions or
trials)
At the level of Procedure (can be design evel to)
SOLUTION:
Counterbalancing of conditions
Use Fixed Sequence: Useful when carryover effects cannot be balanced out or
equated.
Present conditions in a fixed order that works against the hypothesis.
e.g., administer treatment condition first, then control
condition.
Any practice effects will benefit the control condition and work
against the hypothesis.
The sequence becomes a counter-confound.
i.e., not a threat to interpretation, but makes the study less
sensitive (i.e., less likely to find a significant result in the
predicted direction).
Yoked controls: Yoked controls are used in:
developmental research in children (e.g., behaviour
management), operant conditioning in animals (e.g., learning a
button-pressing task), etc..
Provision of rewards has a unique sequencing problem:
The sequence and timing of rewards will differ from participant-
to-participant, because they are contingent on the participant’s
behaviour in response to the manipulations.
Because these important events are under the participant’s
control, they cannot be known in advance, so counterbalancing
is impossible.
Yoked Control – General Procedure:
Yoke participants in pairs and match their condition sequences
(i.e., the sequence of manipulations).
IV: a reward contingent on a target behaviour.
◼ e.g., a rat in a conditioning experiment is rewarded for
pressing a button.
The yoked control receives a reward on the same schedule,
regardless of behaviour.
i.e., the yoked control is rewarded every time
the test participant earns a reward
EXAMPLE:
Brady’s study is a classic yoked control catastrophe!
Monkeys divided into “executives” and controls:
“Executive monkeys” performed a difficult discrimination task under pressure:
Punished by electric shock for errors/delay!!!
Yoked controls had no task to do – they simply received a shock when their
partners were shocked.
Hypothesis: the executive’s pressured decision-making is stressful.
i.e., this stress is not simply the result of aversive events (hence, the yoked
control).
Results:
Found increase in stress symptoms (ulcers) in executive group.
Attributed this to having to make decisions under pressure, not shock per se.
But the monkeys were assigned to groups based on how easily they learned the
discrimination task!
Slow learners became yoked controls (monkey-variable confound!)
Subsequently it was found that faster learners were more anxious and
susceptible to stress.
The executive hypothesis was discredited.
MIGHT HELP IN EXAM:
After being in an angry mood, it may not be possible to put people back into
neutral mood. (They’re just too angry.)
That would lead to diminished differences when angry condition is run first, and
control second.
Ps could get used to the angry face and learn to ignore it, as it is always a
distractor and does not help with the task.
This would lead to diminished effects in 2nd block, with both mood inductions.
Questions:
Is No. 1 a Testing Confound or a Sequence Confound?
Is No. 2 a Testing Confound or a Sequence Confound?
Which of these are possible remedies?
Use a control group and measure DVs concurrently.
Use between-subjects design for the mood conditions and randomly assign
participants to conditions.
take 2 different stimulus sets of faces and counterbalance them across different
mood x order conditions.
Run the conditions on different days.
Make sure that no face is presented more than once.
Use different DVs for the two mood conditions.
Sensitivity
Typically an issue if you get a null result.
Null results are ambiguous:
◼ EITHER:
◼ The hypothesis is incorrect.
◼ i.e., there is no relationship to be found, and
the null result is valid.
◼ OR:
◼ The study was not sufficiently sensitive.
◼ i.e., there is a relationship to be found, but the
study was inadequate to detect it (type 2
error).
Random Variation / Random Error:
Acts as Noise and therefore reduces sensitivity.
Increases error term in statistical analysis.
Renders it less likely that true underlying association will be found (e.g., in
regression, correlation);
Renders it less likely that true underlying differences will be detected (become
statistically significant, e.g., in t-test).
Aim of Experiment:
Find an effect of IV (A causes X).
Measurements usually contain:
◼ Systematic Variation due to Manipulation = Effect [Signal]
◼ Random Error = Unsystematic Variation [Noise]
SOLUTION: Improving Sensitivity
“Increase signal-to-noise ratio”:
Signal: What we want to measure (e.g.: response to treatment).
Noise: Random fluctuations in the response.
Sensitivity can be increased by maximising the IV-DV relationship (i.e., to
increase signal, and improve the signal-to-noise ratio).
◼ To this end, researchers can:
◼ Ensure adequate variation of the IV; e.g.:
(a) Use extreme levels of the IV:
e.g., If attempting to induce fear:
make the manipulation strong (e.g., make the fear-inducing condition very scary
in comparison with the control condition).
e.g., If offering participants an incentive (vs. no incentive):
make the incentive very attractive.
e.g., If studying the effect of practice on “automaticity”:
give participants an extreme amount of practice vs. none.
(b) Use parametric variation:
Varying an IV Parametrically = Stepping systematically through the estimated
effective range of the IV.
Aim: To “Catch the effect” somewhere in the range.
i.e., Helps you to find the most effective manipulation.
e.g., If studying the effect of ostracism on well-being:
Include a large range of different durations of ostracism as levels of the
“ostracism” IV.
The duration that impairs well-being most is optimal
(c) Check the potency of the manipulation.
Tailor the procedures for maximal IV effect.
i.e., Conduct a Manipulation Check on the IV.
e.g. check whether the manipulation had the intended effect.
e.g., Did success/failure feedback affect participants’ beliefs about their
performance?
Could use self-report (e.g., rating-scales): i.e., Was there a substantial difference
in participants’ self-reported beliefs about performance over IV levels?
Or: Did the fear-induction manipulation actually frighten participants?
Assess self-reported fear, or physiological recordings.
Also used for construct validation.
Or to also maximise the IV –DV relationship, you could also tailor the procedures
for maximal IV effect.
e.g., Optimise the level of a non-focal interactive variable (= a factor which is not
an IV in the study, but which has the potential to interact with one of the IVs).
e.g., Use highly-hypnotisable participants in a hypnosis study.
e.g., Use non-drinkers of coffee for a caffeine study in which caffeine-dose is
manipulated.
Because they have lower tolerance to caffeine.
e.g., Use female experimenters in verbal conditioning
Because they elicit more consistent effects.
Note: Trade-off with generalisability.
NOTE: I mentioned that some “CHECKS” can be used in one of two ways.
Document Summary
Exposure to the iv or dv (or associated procedures) causes changes in responding that provide an alternative explanation for the results (e. g. , persistent effects of drug dose, practice, or fatigue) Caused by iv level sequence, not dv measurement. Would change with changes of the sequencing of events (blocked conditions or trials) At the level of procedure (can be design evel to) Use fixed sequence: useful when carryover effects cannot be balanced out or equated. Present conditions in a fixed order that works against the hypothesis. E. g. , administer treatment condition first, then control condition. Any practice effects will benefit the control condition and work against the hypothesis. The sequence becomes a counter-confound. i. e. , not a threat to interpretation, but makes the study less sensitive (i. e. , less likely to find a significant result in the predicted direction). Developmental research in children (e. g. , behaviour management), operant conditioning in animals (e. g. , learning a button-pressing task), etc Provision of rewards has a unique sequencing problem:
