Exploring the Equivalence and Rater Bias in AC Ratings Prof Gert Roodt – Department of Industrial Psychology and People Management, University of Johannesburg Sandra Schlebusch – The Consultants ACSG Conference 17 – 19 March 2010
Presentation Overview Background and Objectives of the Study Research Method Results Discussion and Conclusions Recommendations
BackgroundBackground Construct Validity has long been a Problem in ACs (Jones & Born, 2008) Perhaps the Mental Models that the Raters use are Part of the Problem However, other Factors that Influence Reliability Should not be Neglected
Background Continued To Increase Reliability Focus On all aspects of the Design Model (Schlebusch & Roodt, 2007): Analysis Design Implementation o Context o Participants: o Process Owners (Simulation Administrator; Raters; Role-players)
Background Continued Analysis (International Guidelines, 2009) o Competencies / Dimensions oAlso Characteristics of Dimensions (Jones & Born, 2008) o Situations o Trends/Issues in Organisation o Technology
Background Continued Design of Simulations o Fidelity o Elicit Behaviour o Pilot
Background Continued Implementation o Context: Purpose o Participants o Simulation Administration (Potosky, 2008) Instructions Resources Test Room Conditions
Background Continued Raters Background Characteristics “What are Raters Thinking About When Making Ratings?” (Jones & Born, 2008)
Sources of Rater Bias Rater Differences (background; experience, etc.) Rater Predisposition (attitude; ability; knowledge; skills, etc.) Mental Models
Objective of the Study The Focus of this Study is on Equivalence and Rater Bias in AC Ratings More specifically on: Regional Differences Age Differences Tenure Differences Rater Differences
Participants (Ratees) Region Research Method Region FrequencyPercent Valid Percent Cumulative Percent Valid Western Central Eastern Total
Participants (Ratees) Age Research Method (cont.) Age (Recode) FrequencyPercent Valid Percent Cumulative Percent Valid 30 years or less years years years or older Total MissingSystem Total
Participants (Ratees) Tenure Research Method (cont.) Years of Service (Recode) FrequencyPercent Valid Percent Cumulative Percent Valid 10 years or less years years years or more Total MissingSystem Total
Research Method (cont.) Measurement: In-Basket Test Measuring Six Dimensions: Initiative; Information Gathering; Judgement; Providing Direction; Empowerment; Management Control Overall In-Basket Rating
Research Method (cont.) Procedure: Ratings were Conducted by 3 Raters on 1057 Ratees Observer (Rater) FrequencyPercentValid Percent Cumulative Percent Valid Total
Initiative ResultsResults FrequencyPercentValid PercentCumulative Percent Valid ND R Total
Results (cont.) Initiative Reliability Statistics: Initiative Cronbach's Alpha N of Items.5564 Reliability Statistics: Initiative ObserverCronbach's AlphaN of Items
Results (cont.) Information Gathering FrequencyPercent Valid Percent Cumulative Percent Valid ND R Total
Results (cont.) Information Gathering Reliability Statistics: Information Gathering Cronbach's AlphaN of Items.4853 Reliability Statistics: Information Gathering ObserverCronbach's AlphaN of Items
Results (cont.) Judgement FrequencyPercentValid PercentCumulative Percent Valid ND R E Total
Results (cont.) Judgement Reliability Statistics: Judgement Cronbach's AlphaN of Items.8135 Reliability Statistics: Judgement ObserverCronbach's AlphaN of Items
Results (cont.) Providing Direction FrequencyPercent Valid Percent Cumulative Percent Valid ND R E HE Total
Results (cont.) Providing Direction Reliability Statistics: Providing direction Cronbach's AlphaN of Items.7455 Reliability Statistics: Providing direction ObserverCronbach's AlphaN of Items
Results (cont.) Empowerment FrequencyPercent Valid Percent Cumulative Percent Valid ND R E HE Total
Results (cont.) Empowerment Reliability Statistics: Empowerment Cronbach's AlphaN of Items.7493 Reliability Statistics: Empowerment ObserverCronbach's AlphaN of Items
Results (cont.) Control FrequencyPercent Valid Percent Cumulative Percent Valid ND R E HE Total MissingSystem1.1 Total
Results (cont.) Control Reliability Statistics: Control Cronbach's AlphaN of Items.7485 Reliability Statistics: Control ObserverCronbach's AlphaN of Items
Results (cont.) Overall In-Basket Rating Reliability Statistics: In-basket Cronbach's AlphaN of Items.7686 Reliability Statistics: In-basket ObserverCronbach's AlphaN of Items
Results (cont.) Regional Differences Robust Tests of Equality of Means Statistic(a)df1df2Sig. InitiativeBrown-Forsythe Info GatheringBrown-Forsythe JudgementBrown-Forsythe Providing Direction Brown-Forsythe EmpowermentBrown-Forsythe ControlBrown-Forsythe In-BasketBrown-Forsythe a Asymptotically F distributed.
Results (cont.) Age Differences Robust Tests of Equality of Means Statistic(a)df1df2Sig. InitiativeBrown-Forsythe Info GatheringBrown-Forsythe JudgementBrown-Forsythe Providing Direction Brown-Forsythe EmpowermentBrown-Forsythe ControlBrown-Forsythe In-BasketBrown-Forsythe a Asymptotically F distributed.
Results (cont.)- tenure Tenure differences ANOVA Sum of SquaresdfMean SquareFSig. Initiative Between Groups Within Groups Total Info Gathering Between Groups Within Groups Total Judgement Between Groups Within Groups Total Providing Direction Between Groups Within Groups Total Empowerment Between Groups Within Groups Total Control Between Groups Within Groups Total In-Basket Between Groups Within Groups Total
Results (cont.) Rater Differences Robust Tests of Equality of Means Statistic(a)df1df2Sig. InitiativeBrown-Forsythe Info GatheringBrown-Forsythe JudgementBrown-Forsythe Providing Direction Brown-Forsythe EmpowermentBrown-Forsythe ControlBrown-Forsythe In-BasketBrown-Forsythe a Asymptotically F distributed.
Results (cont.) Post Hoc Tests: Judgement Multiple Comparisons Dependent Variable: Judgement Dunnett T3 (I) Observer (J) Observer Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Lower Bound Upper Bound Lower Bound (*) (*) (*) (*) * The mean difference is significant at the.05 level.
Results (cont.)
Post Hoc Tests: Providing Direction Multiple Comparisons Dependent Variable: Providing Direction Dunnett T3 (I) Observer (J) Observer Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Lower Bound Upper Bound Lower Bound (*) (*) (*) (*) (*) (*) * The mean difference is significant at the.05 level.
Results (cont.)
Post Hoc Tests: Empowerment Multiple Comparisons Dependent Variable: Empowerment Dunnett T3 (I) Observer (J) Observer Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Lower Bound Upper Bound Lower Bound (*) (*) (*) (*) * The mean difference is significant at the.05 level.
Results (cont.)
Post Hoc Tests: Control Multiple Comparisons Dependent Variable: Control Dunnett T3 (I) Observer (J) Observer Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Boun d Lower Bound Upper Bound Lower Bound (*) (*) * The mean difference is significant at the.05 level.
Results (cont.)
Post Hoc Tests: In-Basket Multiple Comparisons Dependent Variable: In-Basket Dunnett T3 (I) Observer (J) Observer Mean Difference (I-J) Std. Error Sig. 95% Confidence Interval Lower Bound Upper Bound Lower Bound Upper Bound Lower Bound (*) (*) * The mean difference is significant at the.05 level.
Results (cont.)
Initiative Info Gathering Judgement Providing Direction Empower ment ControlIn-Basket Initiative1.000 Info Gathering.813(**)1.000 Judgement.448(**).445(**)1.000 Providing Direction.554(**).506(**).493(**)1.000 Empower ment.441(**).428(**).479(**).469(**)1.000 Control.491(**).535(**).419(**).431(**).400(**)1.000 In-Basket.475(**).418(**).761(**).679(**).814(**).595(**)1.000 Non-Parametric Correlations
Clear Regional; Age and Tenure Differences Do Exist among Participants Possible Sources of the Differences: Regional Administration of In-Basket Thus Differences in Administration Medium (Potosky, 2008) o Different Administrators (Explaining Purpose; Giving Instructions; Answering Questions) o Different Resources o Different Test Room Conditions DiscussionDiscussion
Differences Between Participants Regionally: English Language Ability (not tested) Motivation to Participate in the Assessment (not tested) Differences in Employee Selection Processes as well as Training Opportunities (Burroughs et al., 1973) Simulation Fidelity (not tested) Discussion (cont.)
Clear Regional; Age and Tenure Differences Do Exist among Participants Supporting Findings by Burroughs et al. (1973) Age does Significantly Influence AC Performance Participants from Certain Departments Perform Better Discussion (cont.)
Appropriateness of In-Basket for Ratees Level of Complexity Situation Fidelity Recommendations: Ensure Documented Evidence (Analysis Phase in Design Model) Pilot In-Basket on Target Ratees (Design Phase of Design Model) Shared Responsibility of Service Provider and Client Organisation Discussion (cont.)
Context in Which In-Basket Administered Purpose Communicated Recommendations: Ensure Participants (Ratees) and Process Owners Understand and Buy- into Purpose Discussion (cont.)
Consistent Simulation Administration: Instructions Given Consistently Interaction with Administrator Appropriate Resources Available During Administration Test Room Conditions Appropriate for Testing Recommendations: Ensure All Administrators Trained Standardise Test Room Conditions Discussion (cont.)
Rater Differences do Exist Possible Sources of Rater Differences: Background (All from a Psychology Background, with Management Experience) Characteristics such as Personality (Bartels & Doverspike) Owing to Cognitive Load on Raters Owing to Differences in Mental Models (Jones & Born, 2008) Discussion (cont.)
Possible Sources of Rater Differences (cont.): Training o All Received Behaviour Oriented Rater Training o Frame of Reference Different Discussion (cont.)
Recommendations: Frame of Reference Training on: Dimensions, Management-Leadership Behaviour, Norms Project Management Personality Assessment of Raters Sub-dimension Differences
Questions?Questions? ?
SummarySummary Found Rater Bias Need to Research the Source of the Bias Recommend Frame of Reference Training, Project Management Communication of Purpose and Administrator Training