Teaching block 2 - Experimental design - (weeks 4,5,&6) Flashcards
(95 cards)
1
Q
What must research questions be
A
focused
researchable
feasible
specific
complex
2
Q
what is a hypothesis
A
clearly stated explanation based on observations and assumptions.
3
Q
What should a hypothesis do
A
lead to a testable prediction
4
Q
why is quantity not a substitute of quality
A
research takes up time and resources so it is important to have good experimental design
5
Q
why is hypothesis testing reliant on statistical testing
A
statistical tests have specific assumptions that must be met
6
Q
when designing an experiment what needs to be considered
A
treatment structure
design structure
response structure
7
Q
what is the response structure of an experiment
A
the way you will measure response variables
8
Q
in what order should research be carried out
A
observation - question - hypothesis - prediction - test
9
Q
what is an indirect measurement
A
measuring a variable we are not primarily interested in because it is an indicator for the variable of interest
10
Q
why are indicator measurements taken
A
when taking samples of the primary variable is too time consuming, expensive, ecologically damaging…
11
Q
what is an experimental unit
A
smallest individual or object that can be independently assigned to a treatment.
12
Q
give examples of experimental units
A
Person, animal, object that is the subject of the experiment
13
Q
what is a control group
A
Unmanipulated experimental unit
14
Q
what is s negative control, give a bird example
A
manipulated experimental unit that remains unchanged. eg, bird captured but streamer length remains unchanged
15
Q
what is a manipulative experiment
A
Artificially change something about the experimental unit/sample site
16
Q
Manipulative experiment advantages
A
- eliminates confounding factors
- effect size influenced by level of treatment
17
Q
manipulative experiment disadvantages
A
- eliminated effect of biological variation
- damage to ecosystems
- good controls needed
18
Q
correlative experiment advantages
A
- shown natural biological variation
- Easier - less time, effort, work
less animal & organism handling/interference
19
Q
correlative experiment disadvantages
A
- large effect of confounding factors
reverse causation occurs
20
Q
what is reverse causation
A
Instead of A causing B, it’s actually B causing A
21
Q
why are pilot studies important
A
allows collection of preliminary data which determines experimental approaches
22
Q
what is pseudo replication
A
data points not being independent
23
Q
why is pseudo replication an issue
A
independent experimental units is a key assumption of statistical tests
24
Q
what does random variations quantify
A
the extent to which individual samples differ due to things other than what were interested in
25
why should random variation be minimised
to minimise noise and increase precision. And so all variation is caused by factor of interest.
26
how can random variation be minimised
- increasing sample size
- distributing variation between control and treatment groups
- randomisation
27
what should be replicated in a study
- experimental units with in a single treatment
- treatment and control groups
28
advantage of randomisation
- helps achieve independent observations
- minimises natural variation
- reduces bias
29
why is interspersion beneficial
it reduces effect of spatial heterogeneity in environments.
30
what is interspersion
scattering sample units across the environment
31
what is a confounding factor
a variable that influences both the dependent variable and independent variable
32
example of confounding factor using A,B&C
research = effect of A on B, however C also effects B. C is the confounding factor.
33
why is replication important
reduces chance events and outliers effect
captures natural variability
allows estimation of experimental error
increases precision
34
how can the confidence of an experiment be increased
increase sample size
35
what are the 3 types of manipulative experiments
lab
field
greenhouse
36
what are the 2 types of correlative experiments
natural experiments
observational experiments
37
what is the correlational approach
describes existing condition / compares different conditions
38
what is a manipulative lab experiment
the isolated effect of X can be observed
39
advantages of lab experiments
Max control of conditions
confounding factors can be controlled
40
disadvantages of lab experiments
expensive
small sample size
41
advantages of greenhouse experiments
cheep
large sample size
high explanatory power
42
disadvantages of greenhouse experiments
high maintenance
limits space
43
what are the conditions of a greenhouse experiment
moderately controlled conditions eg temp and humidity remain the same
44
field experiment advantages and disadvantage
A = realistic
D = little control (lots of confounding factors)
45
natural correlative experiment advantages
- natural system, natural complexity
- measurable variation despite low experimental control
46
natural correlative experiment disadvantages
confounding factors effect
47
observational correlative experiment advantages
natural system reflects natural complexity
48
observational correlative experiment disadvantages
NO experimental control
confounding factors influence
49
what is an observational experiment
researchers simply observe and record data without interfering
50
what is a natural experiment
"treatment" or change happens naturally, and researchers study its effects
51
which experiment gives the most control
manipulative lab experiments
52
which experiment shows the largest complexity
observational studies
53
as replication increases...
variation decreases
54
what is standard deviation
a measure of spread around the mean
55
if light has no effect on leaf size what should the sample distribution look like
clustered around the same normally distributed mean
56
effect of increasing sample size when an effect is present
estimation of effect size becomes more precise
57
why is effect size important
its needed to determine how many replicates are needed. eg to detect a small effect size a lot of samples are needed
58
why must samples be large
larger samples tend to have smaller standard deviations
59
what must samples be
large
representative
independent
60
what makes a sample independent
if one value does not effect another
61
what does inferential statistics do
uses measurement from a subset of subjects to make a generalisation about the larger population. includes ANOVSA and linear models
62
63
what is pseudo replication
using inferential statistics to test for effects in data that is not independent and replicated
64
what is type 1 sudorepliction
1 unit of replication between control and treatment
several measurements of this 1 unit
these samples are not replicates or independent
65
give an example of type 1 sudorepliction
eg all samples taken from the same plant
66
what is sacrificial pseudo-replication (type 2)
2 replicates pre treatment and control (4 samples). these samples represent the entire population.
better than type one but not good enough to do stats on
67
what is temporal pseudo-replication (type 3)
several measurements taken from sampling unit over time
samples not independent eg value at time one gives a good indication of value at time 2.
68
what is statistical power
the probability that an experiment will detect an effect and the null will be correctly rejected
69
type 1 error
false positive - reject the null when its actually true.
detecting an effect when there is none
70
type 2 error
false negative - accepting null when it should be rejected
not detecting a real effect
71
what is effect size
size of difference between 2 groups/ slope of a relationship
72
what factors affect the power of an experiment
effect size
amount of random variation
number of replicates
73
effect of a large effect size on power
easier detection of difference so increases power
74
why does a large effect size increase power
when effect size large there is a high probability of rejecting the null and the chance of type 2 errors is low, therefore power is increased.
75
what happens to power when effect size is small
large overlap between samples meaning provability of correctly rejecting the null is low. Therefore the experiment has low power
76
what is random variation
the difference in sample units that cannot be explained by the factors of interest
77
what is the effect of high random variation on power
reduces power because the probability of detecting an effect reduces
78
what does random variation do to sample distributions
increases the overlap between distributions
79
how can the effect of random variation be reduced
increasing the number of replicates
80
what must be considered when choosing the number of replicates for a study
large enough to give confidence of detecting meaningful effects
not too large so sampling is un necessary
81
what is a power analysis used for
detects how many replicates are needed to have good statistical power
82
what is a good percentage of power
80%/0.8
83
how can random variation be reduced
improved data collecting techniques (reduces bias)
run a pilot study
84
what type of error does bias create
adds correlated error between sampling units
85
what type of error does inaccuracy (noise) create
adds error that is uncorrelated from one sample to the next
86
how does calibration improve power and what does it involve
checking/setting measuring equipment
- means measurement time is not a confounding factor
87
what does standardising experimental conditions involve - give an example.
taking measurements between specific times of day, temps...
UK butterfly monitoring scheme
88
when should a student T-Test be used
when one factor is being manipulated eg presence/absence of fertiliser
89
What assumptions does the
randomised experiment
balanced experimental design, equal no of experimental units in each group
normally distributed data
90
if the data is not parametric what test shoals be used instead of the student t-test
mann-whitney U test
91
when should the mann-whitney U test be used
one-factor experimental designs with a factor that has 2 levels.
design must be replicated and balanced
92
what statistical test should be used:
randomised 1 factor design with 4 levels
eg effect of 0, 10mg, 20mg, 30mg fertiliser on plant growth
ANOVA - one way analysis of variance
93
what statistical test should be used:
randomised 1 factor design with 4 X levels and 2 Y levels
e.g. effect of 0, 10mg, 20mg, 30mg fertiliser on plant growth and presence/absence of pesticide
Two way ANOVA
94
what is inferential statistics and what is is used for, give examples
uses measurement from a subset of population to make generalisations about the wider population. it can be used to test hypothesis.
it includes ANOVA,T-tests, and linear models
95
what is descriptive statistics, what's it used for, give examples
summarises discrete data and focuses on only the data that has been collected!
mean, median, range and mode
