impact evaluation: balancing rigor with reality
DESCRIPTION
This university lecture at Carleton University shares various evaluation research designs that can be used with community based organizations, especially when a comparison group cannot be identified (i.e. implicit designs and regression discontinuity)TRANSCRIPT
Impact Evaluation:Balancing Rigor with Reality
Donna Smith-Moncrieffe, Public Safety Canada
Carleton University LectureMarch 18, 2014
2
Summary of the Presentation
Background Information National Crime Prevention Strategy Evaluation Strategy
Number of impact and process evaluations per year
Tools in the tool box
Types of project evaluation designs used in the federal government
Statistical tests and examples ANOVA Regression Realist method
Theory of Change Fidelity
- Types of Synthesis Methods
Challenges with balancing rigor and reality
3
Typical # of Evaluations Conducted Each Year
4
Tools in the tool box
5
Types of Designs and Challenges
6
Type of Evaluation Design
Challenge at NCPC, PS
RCT • In government programs, the evaluator is not in a position to randomly assign the youth to a treatment and a comparison group as per the design requirements
• Ethical issues• Contribution agreements are focussed on paying for youth in
treatment groups, not youth in control groups
Quasi-experiment(Delayed comparison group)
• Comparison group is formed with youth on a waiting list where T1 and T2 pre and post tests are completed
• Challenge is that the comparisons at the six month and 12 month post test (T3 & T4) are not possible as the delayed comparison group youth are usually returned to the treatment group
Quasi-experiment(with comparison group)
• Comparison group is formed with youth from a different community where they receive minimal services
• All aspects of internal validity (i.e. maturity, regression to the mean, history etc..) are still a threat unless covariates are assessed to determine how they impacted on the outcome
Types of Designs and Challenges
7
Type of Evaluation Design
Challenge at NCPC, PS
Implicit Design (strengthened version)(Using levels of dosage to create a comparison)
• Identifying the appropriate level of dosage is arbitrary• A validated cut-off for dosage is necessary to make the
appropriate comparison• Not all projects measure dosage in a standardized manner
Regression Discontinuity
• Requires the use of a standardized risk assessment tool that is not readily available or feasible for all projects
• Requires the identification of an assignment variable (i.e. risk level, dosage) that will strongly relate to the outcome variable
• NGOs often want to work with all youth regardless of their level of risk. Program managers may still want youth below the cut-off point to receive the program
Regression Discontinuity Example 1
8
Regression Discontinuity Example #2
9
Impact Evaluation: Using Multivariate Statistics
Table 12.59: T1 and T2 Scores on the Overall Education Attitudes Scale
TIME PERIOD
PIT CLIENTS COMPARISON GROUP N Mean SD N Mean SD
T1 191 53.88 7.23 99 49.51 10.55 T2 191 55.22 5.88 99 49.95 9.96
Sig Testing: The results of a two-way repeated measures ANOVA test shows that the T1-T2-Group interaction effect is not statistically significant (F=1.133, p=.288). Source: T1: B26, T2: A32. Base: Youth respondents with completed T1 and T2 interviews.
10
Impact Evaluation: Using Multivariate Statistics
11
Table 12.61: OLS Regressions Predicting T1-T2 Changes in Education Attitudes
Predictors Model A: ALL YOUTH
Model B: PIT CLIENTS
Model C: PIT CLIENTS
B SE P B SE P B SE P Client (1=PIT Client)
1.532 0.86 .075 -- -- -- -- -- --
Total Dosage -- -- -- .019 0.01 .135 -- -- -- Educ-Related Dosage
-- -- -- -- -- -- .165 0.07 .027
Gender (1=Male) -.843 0.84 .317 -1.087 0.89 .222 -1.457 0.91 .110 Age (in years) -.198 0.15 .184 .085 0.17 .615 .150 0.17 .380 Risk Score .428 0.12 .001 .387 0.19 .009 .419 0.15 .005 Constant -.042 2.75 .988 -4.106 3.13 .192 -5.485 3.22 .090 N 280 181 181 Model Fit: F (p) 3.600 (p =.007) 2.757 (p=.029) 3.466 (p=.009) R Square .050 .059 .073 Adjusted R Square .036 .037 .052
Realist Approach Example
Elements to consider in the Realist Approach
Site 1 (Cree , Quebec) Site 2 (Edmonton) Site 3(Toronto)
Fidelity Rating Low High High
Contextual Differences
• Literacy in English• Parental
engagement is low• Logistical challenges• Cultural differences
• Project implemented as per the model
• Project implemented as per the model
Effect sizes related to emotional regulation (aggression)
• Statistical assumptions not met to report effect sizes
• Moderate to High (0.40-0.61)
• Moderate to High (0.54-1.17)
Cost Benefit Analysis results
N/A Statistical Assumptions not met
1:4 in Total Competencies
N/A Statistical assumptions not met12
Synthesis Methods
13
Types of Synthesis Methods
Systematic Review
• Key points• Summary of rigorous
studies (RCT)• Uses Standardizes
measures (i.e. effect sizes)
• Examples include Cochrane Collaboration, Campbell Collaboration, what works clearing house
•• Challenges:• Narrow focus on program
effects• Does not address
contextual, program factors – how, why questions
• Information about the intervention is absent
Multiple Case Study
• Key Points• Prospective theory-based
approach to synthesis. • Best suited to answer how,
why questions but can also answer what (program effects) questions.
• Open to use of mixed methods (experimental, quasi-experiment , qualitative methodologies).
• Synthesis occurs at the program model level (LST, MST, etc.).
• . Each project = a case.
• Challenges: Development of theoretical framework is a critical first step. Requires extensive coordination between projects to generalize at the program model level.
Realist Synthesis
• Key Points• Retrospective theory-based
approach to synthesis.• Strong focus on what
works for whom, when, and under which conditions.
• Challenges:• Prospective planning is
required but challenging• May not use standardized
measures to synthesize
5
Challenges: Balancing Rigor and Reality
14
Challenges: Balancing Rigor and Reality Cont’d
15
16
Thank You
Contact Information:
Donna Smith-Moncrieffe, BSc., CrimDip, Msc.A/Regional ManagerPublic Safety [email protected]