# Measurement, Part II

Document Sample

```					Measurement, Part II

The challenge

1
Levels of measurement
 Nominal       Ordinal (nonparametric)
 Interval       Ratio (parametric)
 Nominal: lowest level, simply
classifying observations into categories
 Categories should be mutually exclusive
and exhaustive
 Examples: gender, major, religion,
state
2
Levels of measurement
(continued)
 Numbers assigned to the categories
have no numerical meaning. Assign
individuals, and report the % falling into
each category. Fewer statistical
techniques can be used
 Ordinal measurement: one observation
represents more of a given variable
than another observation

3
Levels (continued)
 Rankings
 Newly developed tests
 Ranks tell whether one observation
represents more or less than another,
but not how much more or less--nothing
is known about the exact difference
between any two ranks
 Rankings of crime seriousness

4
Levels (continued)
 Interval: like an ordinal scale, but has
equal intervals between the units of
measurement. Not only an ordering,
but also the same distance or degree of
difference between observations
 For example, 81 is 1 point away from
80, etc.
 Well-developed tests are interval level

5
Levels (continued)
 With interval measurement, can do
division, more statistical tests
 Ratio measurement: like interval, with
the additional property of a true zero.
 an individual could have two or three
time as much of a trait as another with
ratio measurement

6
Levels (continued)
 height or weight. A 200 lb person
weighs twice as much as a 100 pound
person
 Not true for interval. For example, no
such thing as an IQ of 0, and a person
with an IQ of 100 is not twice as smart
as someone with an IQ of 50

7
Levels (continued)
   Most measurement in the social
sciences is interval measurement

8
Developing questions
 Most common method of measuring in
the social sciences is to ask questions
 Types of questions: open ended and
closed ended
 Open: provide own answer. Provides

9
Questions (continued)
 Disadvantages of openended: harder to
 They take longer to answer. Some
questions, and you may end up with
biased results
 Hite reports. Ann Landers

10
Questions (continued)
 Closed ended questions: select an
answer from a list of choices.
 Advantage: quick, easy to code
 Problem: making sure all the possible
reasons are covered

11
 Clear, terms should be defined
questions in one)
 Subjects should be competent to
 Questions should be relevant to the
subjects
 Short

12
Guidelines (continued)
 Avoid negative, or emphasize NOT
 Unbiased
 Recognize social desireability as a
factor when developing questions
(imagine how you would feel giving any
 Use of contingency questions

13
Guidelines (continued)
 Use of matrix questions
 Try to determine if the ordering of the
 Rule of thumb: if questions are written,
questions, and put routine questions at
the end. With interviews, ask routine
first so subjects feel comfortable.

14
Guidelines (continued)
 Always include instructions
 Pretest questions

15
Sampling
 Population: all subjects one is
interested in. Very large or very small
 Element
 Sample: portion of population
 Sampling frame: list of people
(elements) in the populaiton

16
Sampling: continued
 Representative sample: if the overall
characteristics of the sample
approximate the important
characteristics of the population
 Biased sample: not representative
 Parameters and statistics
 Why sample? time and money

17
Sampling in the U.S.
 Literary Digest polls. Accurate until
1936, when Landon was predicted as
winner of the presidential election
 Reasons: (1) low return rates (2 million
out of 10 million) and (2) sampling
frame (telephone directories and lists of
auto owners)
 Poor sampling frames result in bias

18
Sampling in the U.S. (continued)
 1948 Gallup poll predicted Dewey would
win. Problems: (1) stopped polling in
Oct.; (2) quota sampling
 Two types of sampling: probability and
nonprobability sampling
 Probability sampling uses the laws of
probability, whereas nonprobability does
not

19
Probability
 p = number of times an event could
occur / total number of outcomes.
 Can be express as a fraction, a %, as
chances out of 100, or as a decimal.
 P can range from 0 (no probability to 1
(certainty)

20
probability (continued)
 A sample will be more likely to be
representative of a population from
which it is selected if all members of the
population have an equal chance of
being selected in the sample
 Sampling error: error due to the fact
that the sample is not representative
 Necessity of a complete sampling frame

21
Probability sampling methods
 Simple random sampling: (out of a hat,
random numbers)
 Systematic random sampling: every nth
element is cnosen, select first element
at random (random start)
 Stratified random sampling
 1. Divide sample into subgroups based
on important population characteristics
22
P sampling methods (continued)
 2. Randomly sample from those
subgroups in proportion to their
percentage in the population
 Choice of stratification variables will
often depend on what variables are
available, and how much is known
 This technique most likely to be
representative
23
Nonprobability sampling
 Probability sampling only works if there
is a sampling frame of the population.
Sometimes that is not possible (i.e.,
 Nonprobability sampling methods, while
running the risk being unrepresentative
might be the only option

24
Nonprobability sampling
 Convenience: the captive audience
 College students and prisoners
 Purposive: researcher uses judgment
 for Example, the mentally ill. Works
best if the criteria for inclusion are clear.
 Quota: like stratified random. Groups
are selected on the basis of known
variables
25
Nonprobability (continued)
 In quota sampling, however, subjects
are not selected randomly--subjects
with the desired characteristics are
selected until a quota is filled for each
subgroup
 Snowball: each subject is asked to
suggest other subjects

26
 Sample size: unusually the number of
subjects needs to be at least 30. If
several groups within the sample are to
be compared, there needs to be at least
10 per group.
 The larger the number of subjects (N),
the less likely sampling error
 There will always be “mortality”

27
Tips (continued)
 The greater the heterogeneity of the
sample, the larger the sample must be.
The less population diversity, the
smaller N might be.
 N is often determined by time and
money factors

28
Ethics
 No harm to subjects
 Ethics boards or committees, especially
with captive populations such as
prisoners or children (children--parents
must give permission; correctional
systems have their own boards to
protect rights)

29
Ethics (continued)
 Subjects’ right to privacy
 Confidentiality and anonymity
 The only exception: if someone is in
danger
 Voluntary participation
 Informed consent: nature of the study,
possible effects, being able to withdraw
at any time
30
Ethics (continued)
 Deception and debriefing
 Analysis and report: do not “fake”
results, or cheat, or conceal technical
shortcomings of the study
 Milgram’s obedience study
 Zimbardo’s mock prison study

31
Politics and ethics
 Research can be used for good or for
evil
 Ex: development of the atom bomb
 Project Camelot: assessing internal
potential for war, actions governments
might take

32
Politics and ethics (continued)
 Misinterpretation of studies
 Theory of evolution led to social
Darwinism, which led to eugenics and
justifications for Hitler’s purges
 Politics may affect how studies are
interpreted (pornography)

33

```
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
 views: 0 posted: 9/15/2012 language: Unknown pages: 33
How are you planning on using Docstoc?