Flashcards in BIO 330 Deck (379)
Loading flashcards...
91
experimental artifacts
bias resulting from experiment, unnatural conditions
problem w/ experimental studies
should try to mimic natural environment
92
minimum study design requirements
knowledge of initial/natural conditions via preliminary data to ID hypotheses and confounding variables
controls to reduce bias
replication to reduce sampling error
93
study design process
develop clear statement of research question
list possible outcomes
develop experimental plan
check for design problems
94
developing a clear statement of research question
ID question, Ho, Ha
choose factors, response variable
what is being testes? will the experiment actually test this?
95
list possible outcome of experiment
ID sample space
explain how each outcome supports/refutes Ho
consider external risk factors
96
develop experimental plan
based on step 1
outline different experimental designs
check literature for existing/accepted designs
97
develop experimental plan based on step 2
what kind of data will you have- aim for numerical
what type of statistical test will you use
98
minimize bias in experimental plan
control group
randomization
blinding
99
minimize sampling error in experimental plan
replication
balance
blocking
100
types of controls
positive
negative
101
positive control
treatment that should produce obvious, strong effect
ensuring experiment design doesn't block effect
102
negative control
subjects go through all same steps but do not receive treatment- no effect
103
maintaining power with controls
add controls w/o reducing sample size- too many controls samples using up resources will reduce power
104
placebo effect
improvement in condition from psychological effect
105
randomization
breaks correlation btw explanatory variable and confounding variables (averages effects of confounding variables)
106
blinding
conceals from subjects/researchers which treatment was received
prevent conscious/unconscious changes in behaviour
single blind or double blind
107
better chance of IDing treatment effect if
sample error/noise is minimized
108
replication =
smaller SE, tighter CI
109
spacial autocorrelation
each sample is correlated w/ sample area
not independent (unless testing differences in that population)
110
temporal autocorrelation
measurement at one pt in time is directly correlated w/ the one before/after it
111
balance =
small SE, narrow CI
112
blocking
accounts for extraneous variation by putting experimental units that are similar into 'blocks'
only concerned w/ differences within block- differences btw blocks don't matter
lowers noise
113
factorial design
most powerful study design
study multiple treatments and their interactions
equal replication of all combinations of treatment
114
checking for pseudoreplication
check degrees of freedom, very large- problem
overestimate = easier to reject Ho- pretending we have more power than we do
115
determining sample size, plan for
precision, power, data loss
116
determining sample size, wanting precision
want low CI
n ~ 8(sigma/uncertainty)^2
uncertainty is 1/2 CI
117
determining sample size, wanting power
detecting effect/difference
plan for probability of rejecting a false Ho
n~16(sigma/D)^2
D is min. effect size you want to detect
power is 0.8
118
ethics
avoid trivial experiment
collaborate to streamline efforts
substitute models for live animals when possible
keep encounters brief to reduce stress
119
most important in experimental study design
check common design problems
sample size (precision,power,data loss)
get a second opinion
120
