VARIABLES AND MEASUREMENT

1
Lecture 52032005

• VARIABLES AND MEASUREMENT

2
Recap

• Experiments give us control
• Experiments need to be carefully designed to
• eliminate confounds
• ensure the experimental manipulations and data
  give us what we want or what we think we need to
  test the hypotheses and answer the research
  question

3
What is wrong with these studies?
No Control Group spontaneous remission could
account for differences
4
Proof and Disproof

• Given the number of possible things that can/need
  to be controlled little wonder research sometimes
  throws up contradictory results.
• We discussed many of the issues that arise in
  experimental studies last week.

5
Questions about Experimental Design

• Is one study enough?
• Have the variables been properly controlled (e.g.
  order effects, allocation of subjects)?
• Is the sample size large enough?
• Is the sample representative?
• Is the task representative?
• Does the experimental variable have the effect we
  hypothesise it does?
• Is the response variable the correct one to use?

6
The key question is What caused the variation
or differences between experimental groups?

• Was it the sample or the experimental
  manipulation?
• Was the experimental manipulation appropriate?
• Were the correct measures collected and were they
  RELIABLE and VALID?
• Careful theorising and planning are needed for 2
  3 but we use statistics to help with 1

7
The same is true of samples. When samples are
sufficiently different in an experiment, we blame
the experiement not the sampling procedure
The statistical model for defining abnormal
behavior.
The distribution of behavior scores for the
entire population is divided into three sections.
Those individuals with average scores are defined
as normal, and individuals who show extreme
deviation from average are defined as abnormal.
8
From the sampling distribution of the means we
create an expectation and when this is violated
we infer the experimental manipulation caused
the sample to behave as if it came from a
DIFFERENT underlying population.
The population of adult heights forms a normal
distribution.
If you select a sample from this population, you
are most likely to obtain individuals who are
near average in height.
As a result, the scores in the sample typically
will be less variable (spread out) than the
scores in the population.
If we keep sampling we can plot the means to
create a distribution of means
9
Hypothesis Testing

• We dont prove theories or hypotheses we falsify
  them given the data what is the probability
  that the null hypothesis is false? (ie that there
  is no difference between the groups being
  compared?)
• This is a fundamental question that we use
  inferential statistics for.

10
Summary of research design so far

• Define the question
• Decide how you will collect the data
• (Experiment or Field)
• Sample
• Decide what data you will need and from whom
  (Sample, construct measurement)
• Decide how you will analyse the data (Statistical
  Procedure)
• Interpret the results (Inference)
• Communicate

11
Lets move on

• Defining Variables
• Variable Types
• Data properties for analysis

12
Defining Variables

• Developing an Operational Definition
• Easy variables to measure are those that are
  objectifiable (e.g. blood pressure, reaction
  time). They may be measured reliably but are
  they valid measures of the underlying
  psychological construct
• Psychological constructs are often not seen but
  inferred from instruments including
  questionnaires and tests

13
Defining variables II

• The problem for psychological research is what do
  the measures represent (e.g. stress, IQ)?
• This is the leap of faith in many experimental
  studies as the measures are hypothesised to
  reflect hypothetical constructs
• So construct definition and measurement
  validity are vital starting points for research

14
Why is operationalising variables important?

• Facilitates replication
• Promotes understanding
• Promotes critical review

15
Measures vary with respect to what we can do with
them

• Some variables designate identity (male,
  Australian, European, smoker) (Categorical/Nominal
  )
• Others allow you to make transitive inferences
  about magnitude more than, less than (e.g.
  Ranks) (Ordinal)
• Others allow you to make specific statements
  about quantity (temperature, IQ) (Interval)
• Others have an absolute zero (time spent
  studying, number of cigarettes smoked number of
  items recalled in a word list) (Ratio)

16
Scales of Measurement

• Nominal no numerical properties
• Ordinal rank order
• Interval units on the scale are equal but cant
  form ratios (10 degrees Celsius is not half as
  hot as 20 degrees Celsius)
• Ratio has absolute zero, can form ratios (e.g.
  100ms is 2x as long as 50ms)

17
Scales of measurement impact on the information
we can transmit

• We can say least using nominal data frequencies
• Most using ratio scales
• frequencies,
• magnitudes,
• ratios

18
Scales of Measurement impact on the tests we use

• Nominal Chi Square (?2)
• Use assessment of the association between
  different categories of people, things and events.

94 92
TOTAL 58 128
19
Ordinal Scale
Typical tests include Kruskal-Wallis,
Mann-Whitney U Test
20
Interval Scale
Typical Tests t- tests ANOVA Regression
Note assumption of interval equality
How big is the difference?
21
RATIO Scale
Typical Tests t-tests, ANOVA Regression