Title: AND THE INTERNATIONAL DATING VIOLENCE STUDY. Murray A
1 ANALYZING AND INTERPRETING DATA FROM THE REVISED
CONFLICT TACTICS SCALES AND THE INTERNATIONAL
DATING VIOLENCE STUDY Murray A. Straus Family
Research Laboratory, University of New
Hampshire Durham, NH 03824 603-862-2594
murray.straus_at_unh.edu Many papers on the CTS are
available on my website http//pubpages.unh.edu/
mas2 Workshop presented at the International
Conference on Family Violence Research, Portsmouth
, NH 15 July 2003.
2(No Transcript)
3 TOPICS COVERED 1. The multiplicity of CTS
scales and scores and how to manage them2. When
to use the Frequency, Chronicity, Prevalence, and
Ever Prevalence scores3. How to take into
account severity of assault, injury etc4.
Explanation of tables from a paper comparing
Juarez Mexico and New Hampshire, USA5.
Explanation of tables from a paper on the
hypothesis that corporal punishment is related
to violence against a dating partner
4THE FIVE CTS2 SCALES Physical Assault
Injury Sexual Coercion Psychological
Aggression Negotiation Each as subscales for
severe and minor (less severe) levels
5FOUR WAYS OF SCORING EACH CTS SCALE
Prevalence (short for annual prevalence) One
or more occurrences of any of the acts in the
scale Score range 0 - 1 Frequency The number
or occurrences of the acts in a CTS scale Score
range 0 - (25 the number of items in the scale).
For a six item scale this would be
0-150 Meaning of a score of 12 or
25 Chronicity The number of occurrences of the
acts in a CTS scale by a partner who has done
at least one of the acts Meaning of a score of
6, 12, or25 Ever Prevalence One or more
occurrences of any of the acts in the scale since
the relationship began Usually best to avoid
because of memory problems
6DONT LET THE EXTENSIVE DATA FROM THE CTS2TEMPT
YOU TO WRITE AN ARTICLE THAT IS UNDIGESTIABLE
BECAUSE IT HAS TOO MUCH IN IT
- 5 SCALES, EACH WITH 2 OR MORE SUBSCALES, AND
BOTH VICTIMIZATION AND PERPETRATION DATA - 5 2 2 20 variables
- If you include the prevalence rate AND the
chronicity for each scale - 20 2 40 variables
- It is almost always important to present results
for males and females separately, which further
multiples the amount of information in an article - TWO STRATEGIES
- Restrict the paper to either perpetration or
victimization unless you have an important
reason to compare them - Write separate papers for each of the scales
unless you want to compare scores on two scales,
or you have some other important reason to
include two or more
7ISSUES BEST INVESTIGATED WITH EACH OF THE FOUR
SCORES
- PREVALENCE
- What percent of perpetrated or were victimized?
(can also be a rate per 1,000, 10,000 etc) - Among groups in your study such as young and
old, treatment control - In your sample compared with other
populations - FREQUENCY
- How often did the behavior occur in the past
year? (or whatever referent period was used) - Not appropriate with general population
samples because the mean and even the median is
misleading. - Appropriate with batterer treatment
participants or victim samples CHRONICITY - In a general population, among the sub-set who
manifest the behavior, how often did it occur? - EVER PREVALENCE
- If the behavior did not occur in the past year,
was there victimization or perpetration before
that?
8WHY PREVALENCE IS THE MOST FREQUENTLY USED SCORE
- Everyone understands a percentage
- Makes for easy comparison of subgroups
- Does no exacerbate the skewed distribution and
outlier problem - Can be used with logistic regression to obtain
odds ratios and predicted probabilities - Even when other scores are presented, it is
wise to also present the prevalence rates. It
lets readers know where your population is.
9HOW TO TAKE INTO ACCOUNT SEVERITY OF ASSAULT,
INJURY, SEXUAL COERCION, AND PSYCHOLOGICAL
AGGRESSION
- USE THE SEVERE SUBSCALE
- (For Sexual Coercion, the Threat and Force
subscales) - USE THE FREQUENCY OR CHRONICITY SCORE
- Measures severity by how often the behavior
occurs - COMPUTE A SEVERITY WEIGHTED FREQUENCY OR
CHRONICITY SCORE - BUT In a general population sample, exacerbate
skewness and outlier problems - Useful for issues such as measuring change in
severity of perpetration or victimization - USE SEVERITY-LEVEL CATEGORIES
- 0 None, 1 Minor Only, 2 Severe
- Can be the dependent variable in multinomial
logistic regression. This gives separate
coefficients for Minor Only and for Severe - My current recommendation
-
10MODES OF ANALYSIS
FREQUENCY DISTRIBUTION CROSS TAB ANOVA AND
ANCOVA REGRESSION OLS ROBUST LOGISTIC BINARY
AND MULTINOMIAL FACTOR ANALYSIS GENDER SPECIFIC
ANALYSE MULTILEVEL MODELING
11TABLE FROM A PAPER ON VIOLENCE AGAINST DATING
PARTNERS BY UNIVERSITY STUDENTS IN JUAREZ
MEXICOIgnacio Luis RamirezTexas Technological
Universiy
STATISTICAL METHODS USED TO CREATE THE
TABLES INTERPRETING THE TABLES
12EXAMPLE OF TOO MUCH IN ONE ARTICLE
13THE TOP HALF OF THE TABLE
14Two scales is probably as much as should be in
one paper because it is usually important to give
separate results for Severe level (not in the
above table) Both Prevalence and Chronicity (as
above) INTERPRETATION OF DATA IN THIS TABLE
15TABLES FROM A PAPER ON THE HYPOTHESIS THAT
CORPORAL PUNISHMENT IS RELATED TO VIOLENCE
AGAINST A DATING PARTNEREmily Douglas and
Murray A. StrausUniversity of New Hampshire
- WILL DISCUSS
- STATISTICAL METHODS USED TO CREATE THE TABLES
- INTERPRETING THE TABLES
16CTS Rates
These are CTS prevalence rates Gender
differences can be tested using cross tabs and
chi-square. Using ANOVA or ANCOVA lets you
control for other variables and test for
interactions, but with a dichotomous dependent
variable, it violates the assumptions.
17- ANALYSIS OF COVARIANCE
- DEPENDENT VARIABLE
- CTS Injury Prevalence, coded 0-100
- INDEPDENDENT VARIABLES
- Corporal punishment before age 12
- Gender of student
- INTERACTION
- Corporal punishment with Gender
- COVARIATES (controls)
- Age of Student
- Length of Relationship
- Socioeconomic Status of Family
- Social Desirability Scale
- OPTIONS
- Plot
- Effect Size (gives ETA squared)
- This violates assumptions of ANCOVA.
- So does Frequency score because it is so skewed.
Prevalence is the best alternative.
F 4.76, p lt.01
Males
Females
18LOGISTIC REGRESSION IS OFTEN THE BEST APPROACH
- Dependent variable can be dichotomous
(binomial) or multinomial - Dichotomous CTS Prevalence scores (
perpetrating or victimized) - Multinomial examples Severity Level (None,
Minor Only, Severe) - Couple Type (None, Male Only, Female Only,
Both) - Allows for Multiple independent variables and
control variables - Gives thenet effect of each independent
variable - Provides results in two forms that are more
widely understood than regression coefficients - Odds ratio. Example The odds ratio of 1.087
indicates that each increase of one point in the
seven category corporal punishment index
increases the odds of a man physically assaulting
his partner by 8.7 (Straus Yodanis, 1996) - Predicted Probability of the dependent variable
occurring (see example slide) - USE GRAPHS as in Straus, Murray A. and Carrie
L. Yodanis. 1996. "Corporal punishment in
adolescence and physical assaults on spouses
later in life What accounts for the link?"
Journal of Marriage and the Family 58825-841.
19(No Transcript)
20(No Transcript)
21(No Transcript)