120 likes | 297 Views
Introduction. Rapid globalization ? New challengesNational/Regional Cultural IssuesLanguage IssuesIncreased Value of Org. SurveysCentral decision makingOrganizational intervention planningChallenge:Comparing item- and scale-scores across languages and National culture. Background. Large multinational client of Denison ConsultingProblem:
E N D
1. Substantive vs. Quantitative Determination of Item Comparability in an Organizational Culture Survey
Nathan T. Carter
Bowling Green State University
Lindsey M. Kotrba Michael Gillespie
Denison Consulting
Michael J. Zickar Dalia Diab Bing Lin Shuang Pui
Bowling Green State University
2. Introduction Rapid globalization ? New challenges
National/Regional Cultural Issues
Language Issues
Increased Value of Org. Surveys
Central decision making
Organizational intervention planning
Challenge:
Comparing item- and scale-scores across languages and National culture
3. Background Large multinational client of Denison Consulting
Problem: “Can we compare scores on DOCS across language adaptations?”
4. Background BGSU’s Institute for Psychological Research and Application (IPRA) contracted
Item Response Theory Analyses
Differential Item Functioning Analyses
At the same time…
Client conducted focus groups to identify items that were perceived as biased in reference to Reference group (English US)
5. A Brief Introduction to Item Response Theory Why Item Response Theory?
Foundation of contemporary differential item functioning (DIF).
Utilized for examining item-level differences.
Items differ in:
Item Discrimination
Strength of relationship between the trait and responses to the inventory
Option Endorsement Rates
The frequency at which persons endorse a particular item option.
6. The Basic Concept of Differential Item Functioning (DIF) Enables in-depth multinational comparison of self-report inventories at the item level.
Examines score differences between two hypothetical persons who:
1) Took Different Forms of the Inventory
2) Have the Same True Score
Differential Functioning Has Occurred When:
Those with the same True Score have different probabilities in endorsing the item in a particular direction as a function of the form taken.
What Does DIF Imply?
Expected Scores are a function of the form taken.
Scores on the inventory are not directly comparable.
7. Purpose of this Study To examine different trends in the rate of identification of “biased” items
Quantitative Approach (DFIT Study)
Substantive Approach (Focus Groups)
Provide a more thorough understanding of complimentary approaches
E.g. Are they redundant?
8. Results DOCS is generally comparable across language translations
Identification rates significantly higher for Substantive Method (19.2% of 840 item comparisons) than the Quantitative (12.5% of 840 item comparisons)
McNemar c2=15.15, p<.001
Eastern vs. Western languages
Substantive:
Eastern: 31.3%
Western: 14.3%
Kilmogorov-Smirnof z = 2.22, p<.001
DFIT:
Eastern: 15.8%
Western: 11%
K-S z = .63, p = .82
Very little overlap in which items were identified by the methods
Overall – Spearman’s Rho = .08
Eastern – Spearman’s Rho = .05
Western – Spearman’s Rho = .08
9. Results
Translations with most violations (in bold) for quantitative vs. substantive approaches:
None were the same
Decisions would be highly different
Corrections based on Substantive would be far more intensive
10. Results DFIT:
Appears to be more interaction between lingual culture and the construct measured.
11. Results SUBSTANTIVE:
Large differences in identification rates across Regions.
Sizable differences between Regions for all subscales of Organizational Culture
12. Conclusions Unanswered Questions:
No “gold standard” for which is “correct”
Two ways to view this:
1) The Substantive Approach offers unique information
2) The Substantive Approach is biased (Ross & Okabe, 2006)
DFIT procedure
Detection rates less influenced by Region
More “efficient” detection of item bias
Less costly method of item bias detection
13. Recommendations Focus on scores as the primary concern
Decisions made on scores
Quantitative approaches will be more useful in this sense.
Hambleton & de Jong (2006):
“The goal should… be to minimize differences to acceptable semantic, psychometric, linguistic, and psychological levels” (p. 128)
This suggests that meaningful information may be gleaned from both
DFIT clearly gives guidance on psychometric levels
However: Does the substantive approach meaningfully address the semantic, linguistic, and/or psychological levels?
What is known:
Differences do exist in the two approaches
Both have unique identification patterns
More research needed to know the exact benefits and limitations of both approaches
Is one approach “catching” something the other is not?
Are they complimentary?
Is one superior to the other?