week 5 exp designs Flashcards
noise
random variation that disrupts relevant information
what 3 types of noise are there
stimulus precision eg auditory stimulus
participant noise eg motivation
precision of data recording eg buffer of exact time
signal
core element that a person is trying to focus on or interpret.
what does noise decrease
reliability of measurement - more noise = less likely to get same results across measurements
aim of experimental control
maximize the signal relative to noise ratio
2 ways to deal with noise
minimise
neutralise
how to minimize noise
use instruments with high accuracy and precision eg high-end hardware
maximise usability and minimize errors eg make the task easy to understand - minimize ambiguity
maximise ptsp alertness eg take breaks
how to neutralize noise
uniform random variation -When random variation is uniformly distributed, over time or across multiple observations, the effects of noise can average out. This means that the random fluctuations (noise) become less disruptive to the overall pattern or signal, as the randomness tends to balance out. This allows the signal to emerge more clearly, as noise doesn’t systematically bias the outcome in a particular direction.
LARGER DATASET/SAMPLE = LESS NOISE
noise is not absent but neutralized by calculating…
central tendency
how does higher signal to noise ratio affect type 1 and 2 errors
higher signal to noise ratio reduces false negatives (type 2 error, specificity) and false positives (type 1 error, sensitivity) which increases sensitivity and specificity of results
type 2 error or false negative (beta)
when a researcher or experimenter incorrectly concludes that there is no significant effect or relationship between variables, despite the fact that one is present.
4 ways to control confounders
- fix - keep constant
- randomise
- balancing
- measure and model confound
how to fix/keep constant the confounder
make the potential confounder a control variable eg lab exp
limitations of fixing/keeping confounder constant
fixing is not always possible - order effects
benefit of randomize (neutralize) a confounder
converts confounder (systematic variation) into noise (uniform random variation) - becomes unrelated to IV
any random variation can be neutralized by calculating central tendency now
limitations of using balancing to reduce confounder
only works if you know confounder beforehand
disadvantage of using randomization to reduce confounder
must have sufficient sample size
difference between noise and a confounder
Confounder: A variable that causes a distortion in the relationship between the independent and dependent variables, often leading to incorrect conclusions about causality.
Noise: Random, unsystematic variations or errors in the data that obscure the signal or relationship without creating any false causal associations
how to balance confounder variable
dividing set into 2 or more subsets with roughly the same characteristics
equalize central tendency across conditions so confounder on DV is neutralized
values of the confounder are distributed in equal frequencies across conditions so central tendency is unbiased
what does measure and model mean in terms of reducing confounder
when you can’t fix or neutralize a potential bias, include It when analyzing data
confounder can become an IV - results indicate potential biases
bias
unwanted signal producing spurious results
2 types of bias
independent and dependent bias
independent bias
= systematic errors
systematic tendencies in measurements that are incorrect, but limited to one variable
bias does not vary with the independent variable so can’t be mistaken for causal
dependent bias
= confounder
produces bias in DV, that varies with IV. so interferes with results as produces unwanted, spurious results
noise and systematic errors reduce…
reliability and accuracy
confounders reduce…
internal and external validity
how to minimise systematic errors
calibration - comparison and adjustment to reference measure
what is the smaller unit of an experiment
trial - one completion of the task providing one datapoint of DV
what can you organize trials in
blocks - specific characteristics concerning the control of confounders and manipulation of IV
2 major approaches to manipulating IV
repeated measures design aka within-subjects, paired samples
between-groups design aka between-subject
repeated measures design
all ptsp in all conditions - everyone does everything but IV is manipulated within ptsp
2 ways to organize sequences of repeated measures - blocked vs interleaved
strengths of using repeated measures
reduces noise and cofounds by excluding differences between group
wanted when practice effects are desirable
block design in repeated measures
values of iv in separate blocks
participants are exposed to all conditions multiple times, and these conditions are grouped into blocks.
cons of using block design
order effects
how to deal with order effects
counterbalancing
interleaved design in repeated measures
Different tasks, stimuli, or conditions are presented in an alternating or random order.
pros of using interleaving method
order effects are controlled through randomization
cons of using interleaved design
noise due to unexpected changes
interaction effects - bias when effects of sequence differ between conditions
pros between-groups design
controls order effects
wanted when inexperienced, naive ptsp needed
cons of between-groups design
potential noise and confounds because groups vary due to ptsp characteristic
large sample needed to allow for neutralizing by randomisation
doesn’t control for important ptsp characteristics
vulnerable to drop out - if one pair drops can’t use the other
2 types of between groups design
random groups
matched pair design
what does factorial designs mean
multiple independent variables that can have interaction between factors (iv)
2 types of factorial design
manipulation of factors
mixed factorial designs
pros of factorial designs
internal validity - more than one hypothesis can be tested
allows for assessing complex causes relationships eg interaction effects
allows for including confounders
ecological validity - closer to real world
quantitative interaction effects
The effect changes in size but not direction (e.g., caffeine improves reaction times more in the morning than in the afternoon, but it always helps).
qualitative interaction effects
The effect changes in direction depending on the level of the other IV (e.g., caffeine improves reaction times in the morning but worsens them in the afternoon).
multivariate designs
more than one DV
criticisms of little Albert study
no attempts at removing the conditioning after, negative consequences on development
what was the monster study
assessment on psychological support on stuttering
22 children from iowa soldiers orphan
half received negative treatment eg belittling and half positive
ethical issues of the monster study
no consent, debriefing or help after with negative psychological effects eg speech problems for rest of their lives
ethical issues of milgram shock exp
deception, no debrief, inflicted insight = pushing ptsp to fave negative characteristics of themselves
ethical issues of SPE
no consent, suffered psychological and physical har, hindered right to withdraw and debriefing too late
ethical issues of Harlows monkeys
unethical treatment of animal which harmed them beyond exp
what does the Belmont report include
informed consent, voluntary participation, no/minimal deception, anonymous data, confidentiality, not harmful, fair selection of ptsp
3 factors of ethics
- informed consent
using participation information sheet and gaining consent after all info given - debrief
- review
application to ethics committee and approval necessary
main problem of independent sample designs
ptsp are different - potential confounders
how can u visually examine if a distribution of a variable is biased
with a histogram
statistical population depends on the…
hypothesis
clever Hans was an example of
primary observer effect - horse was responding to involuntary cues in the body language of human trainer
