SCRM Text 7 Content analysis and dictionaries

Traditions of Text Analysis
Text as Proxy for Experience
Structured
Interviews
Analysis of:
List data
free lists
Relational data
pile sorts, triad tests
Combination
Successive free lists
Multidimensional Scaling
Cluster Analysis
Correspondence Analysis
Text as Object of Analysis
Free-flowing
Text
Analysis of:
Analysis of:
Conversation
Performance
Words
KWIC
Word Frequency
Semantic Networks
Narratives
Grammatical structure
Codes
Exploratory
Confirmatory
Grounded Theory
Schema Analysis
Content Analysis
Content Dictionaries
Adapted from Gery Ryan and H. Russell Bernard (2000)
Typical Process of Content Analysis
Theory and rationale
Conceptualizations
Operationalizations
Human coding
Computer coding
Coding schemes
Codebook
Coding form
Coding schemes
Dictionaries
Training and pilot reliability
Human coding
! 2 coders
Sampling
Computer coding
Coding
! 10% overlap
Coding
Final reliability
Tabulation and reporting
Content Analysis Example
(Testing market exchange theory on personal advertisements)
Goal/Hypotheses: Men and women seek complementary qualities in personal advertisements
Data: Personal advertisements from magazines
Resource
[Hirschman 1987]
Hypotheses
Men
Women
Physical Status
Seek
Offer
Money
Education
Occupational
Intellectual
Offer
Offer
Offer
Offer
Seek
Seek
Seek
Seek
Love
Entertainment
Seek
Seek
Offer
Offer
Demographic
Ethnic Info
Personality
Seek
"
"
Offer
"
"
Content Analysis Example
(Data Collection & Analysis Steps)
• Select data source
The Washington and the New York Magazines
• Pretest to assure that resource categories are exhaustive and mutually exclusive
– Identified 100 resource items from a random sample of 20 advertisements
– 11 women & 10 men sorted all 100 items into categories
• Sample
– 405 randomly selected advertisements (~100 males & ~100 females / magazine)
– 3782 resource items
• Intercoder agreement
–
–
–
–
2 coders (male & female)
Omissions: 480 (12.7%)
Discrepancies: 636 (16.8%)
Discrepancies decided by 3rd coder
• Test for differences across groups
[Hirschman 1987]
Content Analysis Example
(Content Coding Steps)
Example:
Single, tall, blond, athletic male ISO beautiful, rich, divorced, female,
interested in tennis.
Resource items:
Offers: 1) Single, 2) Tall, 3) Blond, 4) Athletic
Seeks: 5) Beautiful, 6) Rich, 7) Divorced, 8) Interested in tennis
Step 2
Step 1
ID
1
2
3
4
5
6
7
Verbatim
Offers/
Resource Sex Seeks
Single
M
O
Tall
M
O
Blond
M
O
Athletic
M
O
Beautiful
M
S
Rich
M
S
Divorced
M
S
Interested M
S
in tennis
8
...
3782 Humorous
F
S
Step 3
Resource Category
Coder Coder
1
2
Agree
Dem
Dem
Yes
Phys
Phys
Yes
Phys
Phys
Yes
Phys
No
Phys
Phys
Yes
$
$
Yes
Dem
Dem
Yes
Coder Final
3 Category
Dem
Phys
Phys
Phys Phys
Phys
$
Dem
Enter
Occup
No
Enter
Pers
Pers
Yes
Enter
Pers
Content Analysis Example
(Results)
Hypotheses
Men
Women
Confirmation
Men
Women
Physical Status
Seek
Offer
Seek
Offer
Money
Education
Occupational
Intellectual
Offer
Offer
Offer
Offer
Seek
Seek
Seek
Seek
Offer
ns
ns
ns
Seek
ns
ns
ns
Love
Entertainment
Seek
Seek
Offer
Offer
ns
ns
ns
ns
Demographic
Ethnic Info
Personality
Seek
"
"
Offer
"
"
ns
"
"
Offer
"
"
Resource
[Hirschman 1987]
Race and Ethnicity in Nursing Research
No. of articles :: % with race/ethnicity
90
80
70
60
50
40
30
20
10
0
1952 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
Source: Drevdahl et al. (2001) Nursing Research 50:305-13
Research Questions
How often and in what context do medical
anthropologists use the concepts of race or ethnicity?
Research Questions
How, if at all, do they distinguish
between these concepts?
Research Questions
How often do medical anthropologists explicitly
identify racism and social inequalities
as causes of health disparities?
Codebook Development
Sampling Design
Med Anthropol
(1977-2002)
Med Anthropol Q
(1987-2002)
Sampling Frame
N = 843
Systematic
Random Sample
N = 422
Empirical
Research Articles
60
40
20
0
Percent of articles
80
Frequency of Race and Ethnicity in Health Research
1975
1980
1985
1990
1995
2000
Year
Medical Anthropology
Am J Public Health, 1996-99
Am J Epidemiol, 1975-90
Am J Public Health, 1980-89
Am J Epidemiol, 1996-99
Health Services Research
Gravlee, C. C. & E. Sweet (2008). Med Anthropol Q 22(1):27-51
Race versus Ethnicity in Medical Anthropology
Gravlee, C. C. & E. Sweet (2008). Med Anthropol Q 22(1):27-51
Percent of Articles Using Race or Ethnicity,
U.S. versus Non-U.S. Research
Region
No race or ethnicity
Race alone
Ethnicity alone
Race and ethnicity
Total
Non-U.S.
(n = 178)
U.S.
(n = 105 )
Total
(n = 283)
74.7
54.3
67.1
2.8
7.6
4.6
18.5
19.1
18.7
3.9
19.1
9.5
100.0
100.0
100.0
Two of 93 articles (2.2%) that used race or ethnicity defined these terms
Gravlee, C. C. & E. Sweet (2008). Med Anthropol Q 22(1):27-51
Mention of racism or inequalities
When medical anthropologists talk about race,
they are likely also to talk about racism or social inequalities
No
79.4%
79.4
38.5%
38.5
20.6%
20.6
No
61.5%
61.5
Yes
Yes
Uses any race concept
!2 = 32.11, p < .001
Content Dictionaries
Content Dictionaries
Use computers to code fixed units of analysis
Increase reliability over human coders
Less time-intensive
Types of Dictionaries
All-purpose
Specialized
Individual
All-purpose
General Inquirer
Specialized
Gottschalk-Gleser psychological scales
Individual
Created for a single study (e.g., Colby 1966)
Marking of Text is Automated
• Computer reads the files and ignores a
stop list
• Cuts off suffixes to create word stems
• Uses algorithm to assign each stem to a
particular category
• Usually 80-90% of text is tagged; human
coders review and code untagged terms
Content Dictionary Example
General Inquirer
• Uses the Harvard Psychosocial Dictionary IV
• 8500 entries
• Distinguishes among multiple meanings of
words
Content Dictionary Example
General Inquirer
• Ogilvie et al. (1966) tested General Inquirer
on 66 suicide notes—33 real, 33 simulated
• Men were matched on age, occupation,
religion, and ethnicity
• General Inquirer picked actual suicide notes
91% of the time
Specialized Dictionary Example
(Difference between Navaho & Zuni perceptions of home)
Data: Thematic apperception tests with Navaho and Zuni respondents.
Observed & Hypothesized Relationships:
H1: Navaho regarded their homes as havens and places of relaxation.
H2: Zuni described their homes as places of discord and tension.
Colby and colleague had previously developed two word groups or dictionaries.
Relaxation
assist
comfort
easy
affection
happy
play
[Colby 1966]
Tension
destruction
discomfort
difficult
dislike
sad
battle
anger
Specialized Dictionary Example
(Testing of hypothesis)
Data to test hypothesis:
35 sentences that had the word home and one of the words in the two groups.
Tension and home
Relaxation and home
Navaho
6
Zuni
10
13
6
Conclusions:
Navahos were twice as likely to use words from the relaxation group when
talking about home than they were from the tension group.
Zuni almost twice as likely to use tension words than they were to use
relaxation words. The results were significant at the .10 level.
[Colby 1966]