Biology 1413-151-152-451
Lecture # 03A,
Scientific Method continued
Statistics and Relative Risk
Some say “Figures can’t lie, but liars can figure”,
Others argue “Figures can lie, and liars do figure”…
Learning the Basics, Sample Question:
1) How does a double-blind experiment
decrease the amount of bias introduced into
experimental results:
A double-blind experiment ensures:
that subjects do not influence the results by
reporting what they think is expected based
on the hypothesis,
and it ensures that data collectors do not
err in the measurement of results by
emphasizing supportive data and minimizing
non-supportive data.
Learning the Basics, Sample Question
2) If the results of an experiment are exactly
what was predicted by the hypothesis, then:
a) the hypothesis is proved.
b) the alternative hypotheses are falsified.
c) the hypothesis is supported.
d) the hypothesis was scientific.
e) none of the above.
Answer on next slide
Learning the Basics, Sample Question
2) If the results of an experiment are exactly
what was predicted by the hypothesis, then:
a) the hypothesis is proved.
b) the alternative hypotheses are falsified.
c) the hypothesis is supported.
d) the hypothesis was scientific.
e) none of the above.
The Scientific Method:
Analyze the Data
Understanding Statistics
Statistical Significance isn’t foolproof:
does NOT tell us if the experiment was designed
and carried out correctly
does NOT rule out possibility of sampling error
does NOT rule out alternative hypotheses
also, in the case of Correlation Studies, even if
the correlation is statistically significant this
does NOT necessarily mean causation
Using Correlation to Test Hypotheses
1) Especially used in human experiments
when difficult to control all factors, or
if it’s unethical to give a treatment, or
unethical to not give treatment
2) There is no Control Group, so correlation
does not control well for alternative
hypotheses, so may or may not
imply cause and effect
3) Can be useful, but usually must be followed
by additional testing
Correlations Essay format test
Dependent variable on Y
Degree of relationship
(co-relation)
between two variables
can be calculated
mathematically
Independent variable on X
Stronger the correlation,
closer the relationship
(not just two random variables)
Correlations True/ False format test
Degree of relationship
Dependent variable on Y
(co-relation)
between two variables
can be calculated
mathematically
Independent variable on X
Stronger the correlation,
closer the relationship
(not just two random
variables) NEGATIVE
lower
decreases success.
Analysis: Difficulties with Cause & Effect
Some studies suggest a correlation between
high psychological stress and rates of illness
with common cold
But does high stress cause high frequency of
illness with colds,
or does one of the causes of
high stress also cause
high frequency of illness
with colds?
(alternative hypotheses)
Difficulties with Cause & Effect
Examples:
wearing a necktie is correlated with executive
position, but wearing the tie does not cause
promotion to executive position
having red/blond hair is correlated with
increased risk of sunburn, but having red hair
does not cause the increased risk of sunburn
Other examples of Analysis Difficulties:
Is it correlation or is it cause and effect?
Does eating one serving of Special K cereal each
day cause weight loss, or is it only correlated
with weight loss?
More than one possible hypothesis :
Weight loss is caused by special characteristics
of cereal,
or is it due to eating limited calories
for breakfast (and low fat food)?
Analysis: Difficulties with Cause & Effect
High correlation between cold outdoor
temperatures and increase incidence of
common cold infection
But does cold outdoor temperature cause the
increased incidence of cold infection,
or does increased exposure to viruses due to close
contact with other people cause increase incidence
of cold infection?
Difficulties with Cause & Effect
Having high blood pressure is positively
correlated (associated) with higher risk of heart
attack and stroke, but it’s not “the 1⁰ cause” of
heart attack or stroke (appears to be
contributing, but not a single, direct cause)
If it did, then everyone with high blood pressure
would have heart attacks and strokes, (which is
false) and
conversely- those without high blood
pressure would never have heart attacks
and strokes (which is also false)
Learning the Basics,
Sample Question
3) What are the advantages and disadvantages of
using correlations to test hypotheses?
The advantage is that a correlation allows
researchers to test hypotheses that are difficult to test
via controlled experiments by looking for a relationship
(association or correlation) between two factors.
They have the disadvantage of lacking controls,
meaning that not all alternative hypotheses for the
relationship can be excluded. This disadvantage is
one reason why correlation does not necessarily
mean causation.
The Process of Science:
Understanding Statistics
Statistical significance versus biological
significance or practical significance or social
significance….
Not always the same level of meaning
Secondary news sources often do not distinguish
between statistical significance and more casual
use of the word significance (“meaningful
or important”)
Is it biologically significant? Does it matter in life, for how many?
Is the cost worth the benefit? Does it help only a few?
Should it be available to everyone? Who pays the cost?
Example of a Box-and-whiskers graph
measures
1000
700
500
Control
Group or
Baseline
Test Group #1 Test Group #2
n=50
n=25 n=25
treatment
Example of a Box-and-whiskers graph
This is what’s called a Bell Curve,
1000
or a Normal Curve, (here it’s green
with a red top) and shows how the
distribution of data points are
700 clustered by frequency– how
similar each data point is to the
500 arithmetic mean or the “average”
And 95% of the data points fall
inside the green area of the Curve.
95% of results are in this grey-box or
green-colored bell curve
Each Grey Box is like a green-colored “bell curve”
turned on it’s side, and 95% of the “measurements” in Outliers, or Outliers, or
sample falling inside the box. Only the “outliers” or just “due to just “due to
chance” chance”
“just due to chance data” fall at the Edges of the Curve
(like the Whiskers on the Box-and-Whiskers graph)
The 95% Confidence Interval
As long as the mean (average) of one sample
does not overlap with the confidence interval
for the other sample (or vise versa),
then the average difference between the
two means (two averages) is
statistically significant.
Box-and-whiskers Graphs
Whiskers show range of values in the data from
this sample or group
Cross-hatch shows average (arithmetic mean) for
the data from sample or group, not always “even”
Shaded box encloses 95% confidence interval
The whole sample (or all your data) has a 95%
likelihood of being within the box on the graph.
Bell Curve with
Normal
Distribution
of Data Points
Graphs of Data Point Distributions
Fall into “Bell Shapes”
The 95% Confidence Interval
If the mean (average) of one sample
does overlap with the confidence interval
for the other sample (or vise versa),
then the average difference between
the two means (two averages) is
not statistically significant.
See next slide for the soap example.
Sample Box-and-whiskers graph
1
Dependent variable on Y
1000 ……………………………...…
b 3
700 ……………………………….
2
500 ………………
Independent variable on X
Groups 1 and 2, and 1 and 3 are significantly different (statistically), whereas notice
Groups 2 and 3 may “show a trend” but are not significantly different (statistically). See
how the average value (mean) of Group 2 overlaps with Confidence interval (grey box)
for Group 3 (see arrow “a”) and vise versa (see arrow “b”).
The 95% Confidence Interval
If the overlap between mean (average) of
one sample with the confidence interval of
the other sample is small,
then there is a suggestion of a difference, and would
need to investigate further to find out if really
a meaningful finding.
So “washing with soap was better than before
washing”, but whether subjects used anti-bacterial
soap versus regular soap did not make a statistically
significant difference in number of bacterial colonies–
only a suggestion of a difference.
The Process of Science:
Understanding Statistics
Statistical significance versus biological
significance or practical significance or social
significance….
Not always the same level of meaning
Secondary news sources often do not distinguish
between statistical significance and more casual
use of the word significance (“meaningful
or important” or “draws audiences and makes $$”)