Академический Документы
Профессиональный Документы
Культура Документы
Questions,
Variables, and
Hypotheses
2/21/06
1
Review
What are research questions?
What are variables?
Definition
Function
Measurement Scale
2
Hypotheses
OK,now that we know how to set
up a research project by posing
research questions and labeling
variables, it is time to move on to a
more formal way of structuring and
interpreting research.
3
Hypotheses Definitions
Hypotheses are predictions about the
relationship among two or more
variables or groups based on a theory
or previous research (Pittenger, 2003)
Hypotheses are assumptions or
theories that a researcher makes and
tests.
Why are hypotheses important?
4
Importance of Hypotheses
Hypotheses:
Direct our observations
Identifies the variables examined and data to
be collected
Describe a relationship among variables
Can state that as one variable increases, the
other will decrease; as one variables
increases, the other will increase, and so on.
Refer to populations
Hypotheses help researchers infer that
results of a sample will translate to a
population
5
4 Functions of Hypotheses
Hypotheses can:
Estimate Population Characteristics
Correlate Variables
8
Research Hypotheses
Research Hypothesis: A statement of
the relationship among two or more
variables or groups.
The acceptance or non-acceptance of
which is based on resolving a logical
alternative with a null hypothesis.
12
Statistical Hypotheses
Statistical Hypotheses are mathematical, or logical
statements that help researchers interpret the
results of research
Null: (H0: 1 - 2 = 0 )
Alternative: (H1: 1 - 2 0)
13
Written out examples:
Null: There will be no difference in the
15
Testing Hypotheses
Isit possible that our observations
occurred by chance?
16
A story(borrowed from
Gonick & Smith, 1993)
In the southern US several legal cases
examined racial bias in jury selection.
18
Testing Hypotheses Cont.
22
Error Types Cont.
The picture on the next slide,
which is modified from Hays (1986)
and Ferguson and Takane (1989),
illustrates the relationships among
Type I and Type II errors and alpha
and beta.
23
Error Types Cont.
Error Types Chart
H0 is H1 is
True True
25
26
Power, Effect Size, and
Measurement
STATISTICAL POWER
So, if you have a 20 percent chance of being found not guilty, even
though you committed the crime, the statistical power of the
judicial system is:
1 - .20 = .80.
27
Power
Statistical power is a function of the preset
significance criterion [alpha], the reliability of
sample results, and the effect size [the actual
size of the difference or strength of the
relationship]... (Cohen, 1988, p. 4).
33
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH
DESIGN, ANDSTATISTICAL POWER
34
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH DESIGN,
ANDSTATISTICAL POWER Cont.
Consider the elephant fable with the researchers who
mapped different parts of the elephant. Their descriptions of
the elephant differed considerably.
35
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH
DESIGN, ANDSTATISTICAL POWER Cont.
Now, consider measuring the same elephant with
portable X-Ray machines. Pictures of each part of the
elephant are taken and then compared with each other.
Not only do these pictures not resemble each other,
but they also dont resemble the descriptions provided
by the previous group of researchers. This chapter of
the elephant fable indicates how what we see is
indicated by our method of observation or
measurement.
Again, a researcher interested in a deeper
understanding of the elephant may choose multiple
methods of measurement in order to avoid threats to
validity from mono-method bias.
36
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH
DESIGN, ANDSTATISTICAL POWER Cont.
The relationship of measurement, research design, and
statistical power means that large treatment effects can
actually be observed as small effects. In other words, even
if an intervention is very effective, measurement and
design complications may make the effect appear small and
thus require high statistical power for detection.
37
Power, Effect Size, and
Measurement Cont.
Telescope Model
Actual Effect Size Observed Effect Size
Statistical Design
Research Design
Measurement
38
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH
DESIGN, ANDSTATISTICAL POWER Cont.
Apparent effect size is further attenuated when
research design does not fully filter out extraneous
sources of variation (e.g., counselor or client
differences). Violations of assumptions of statistical
procedures can further attenuate effect size.
Interestingly, problems in research design and
statistical design can also introduce sources of Type 1
error (e.g., dust on the lens or false positive results).
The relationship of effect size, measurement, and
design is further complicated by the frame of reference
or angle from which one approaches or operationalizes
the construct. E.g. the researcher that only looks at
the leg of the elephant.
39
Power, Effect Size, and
Measurement Cont.
RELATIONSHIP OF MEASUREMENT, RESEARCH
DESIGN, ANDSTATISTICAL POWER Cont.
40
Power, Effect Size, and
Measurement Cont.
PRE-ANALYSIS STATISTICAL POWER
42
Power, Effect Size, and
Measurement Cont.
ALPHA INFLATION
Multiple comparisons can increase
alpha, the probability of a Type I error.
45
Test Statistics,
Probability, and
Significance
In order to test a hypothesis, we
compare the obtained value of a test
statistic (e.g., the obtained F) to a
critical value of the test statistic (e.g., a
critical F) that is associated with the
preset significance level (alpha).
If the obtained value of the test statistic
is greater than the critical value, we
determine that there is a significant
difference or relationship.
46
Test Statistics,
Probability, and
Significance
Test Statistic: TheCont.
specific statistic (i.e., the
tool) that is chosen to test the null hypothesis.
Examples include F, t, r.
Obtained Value: The actual value obtained
when applying the test statistic to the data of
interest. The probability value associated with
the obtained value is p.
Critical Value: The critical value of the test
statistic that is associated with the chosen
significance level (alpha). If the obtained value is
greater that the critical value, the result is
significant.
47
Test Statistics,
Probability, and
Significance Cont.
Probability Value: The probability that observed
relationships or differences are due to chance.
Alpha: Alpha is also known as significance level or
rejection region. It is the level of probability set by the
researcher as grounds for rejection of the null
hypothesis (Williams, 1986, p. 58). Alpha is the
probability level associated with the critical value of the
test statistic.
In other words, alpha is our predetermined risk that
differences that we declare to be real are actually due
to chance.
Obtained: This is also known as the obtained
probability (p): significance of the test statistic . It is
the probability that the data could have arisen if Ho
were true (Cohen, 1994, p. 998).
48
Test Statistics,
Probability, and
Significance
Cont.
Significance: What happens when the obtained
probability p is less than our predetermined
alpha. Significance also occurs when the obtained
value of the test statistic is greater than the
critical value of the test statistic.
Test Statistic Probability Value
Critical Value Significance Level (alpha)
Obtained Value Obtained or Actual Probability (p)
Note that larger obtained values of test statistics are
generally related with smaller values of p.
If Obtained Value > Critical Value, then * Significance *
If p < Alpha, then * Significance *
49
Revisiting the Bell Curve
50
Test Statistics,
Probability, and
Significance Cont.
Whether you are looking at obtained
values of test statistics in relation to
critical values or you are looking at
actual probability levels, it is important
to note that the test statistics and their
associated probabilities only tell us the
probability that a difference or
relationship occurred by chance
52
Phase I: Research
Hypotheses, Design, and
Variables
1. State your research hypotheses.
2. Decide on a research design based on your
research problem, your hypotheses, and
what you really want to be able to say about
your results (e.g., if you want to say that A
caused B, you will need an experimental or
time-series design; if probable cause is
sufficient, a quasi-experimental design would
be appropriate).
3. Operationally define your variables. Recall
that one variable can have more than one
operational definition.
53
Phase II: Statistical
Hypotheses
1. Consider your chosen statistical
procedures.
2. Write one statistical null
hypotheses for each operational
definition of each variable that
reflects that statistical operations
to be performed.
54
Phase III: Hypotheses
Testing
Complete the following steps for each statistical null
hypothesis:
1. Compare the obtained value of the test statistics with the critical
value associated with the selected significance level or compare
the obtained p-value with the pre-selected alpha value.
1. If the obtained value of the test statistic is greater than the critical
value (or if the obtained p-value is less than the pre-selected alpha
value), reject the null hypothesis. If the obtained value is less than
the critical value of the test hypothesis, fail to reject the null
hypothesis.
56
Points to Consider about
Hypotheses Testing
FISHING IN LAKE ALICE
We dont prove the null
hypothesis.
If you go fishing on Lake Alice and
you dont catch fish, you cannot
conclude that there are no fish in the
lake!!!
57
Points to Consider about
Hypotheses Testing Cont.
What circumstances might keep us
from finding fish in the lake? Possible
problems include:
1. wrong or insensitive outcome measures
(using the large net for small fish),
2. sampling problems (looking in the wrong
part of the lake), or
3. methodological problems (scooping the
top of the lake rather than the bottom,
where the fish hang out).
58
Points to Consider about
Hypotheses Testing Cont.
Returning to hypothesis testing:
Failure to reject the null hypothesis cannot
be interpreted as proof that no differences
or relationships exist. Existing differences
or relationships might be obscured by:
1. insensitive outcome measures (the wrong
fishnet),
2. inappropriate statistical designs,
3. poor sampling strategies, and
4. low statistical power.
59
Questions about
Hypotheses
60