Canadian Evaluation Society presentation-2011_ final ppt
Using a Fidelity Index to Increase Program Attribution
1. Using a Fidelity Index to Increase Program
Attribution in Impact Evaluation Studies
Presented by Donna Smith-Moncrieffe
Canadian Evaluation Society-Ontario Chapter
October 4th-5th, 2010
2. Presentation Outline
National Crime Prevention Centre (NCPC) mandate
What is a Quantifiable Fidelity Index? Why is it Important?
How can using a Quantifiable index increase Program
Attribution?
Steps: From Fidelity Tool Development to interpretations
regarding program attribution
Sample Fidelity tool and output data for two Model programs
Towards No Drug Program (results)
Stop Now and Plan Program ® (application of treatment intensity
measures)
Summary
2
3. NCPC Mandate/Core Activities
Mission statement:
To provide national leadership on effective and cost-efficient
ways to both prevent and reduce crime by addressing known
risk factors in high risk populations and places
Core activities:
Supporting targeted interventions in local communities
Building and sharing practical knowledge with policy
makers and practitioners
3
4. NCPC Priorities:
Provide funding to the following target groups/crime issues:
Children and youth at risk
Crime prevention in Aboriginal communities
Prevent recidivism among high-risk groups
Priority crime issues (youth gang, drug-related crimes)
4
5. Use of Evaluation in NCPC
Disseminate results
and encourage
the province and
NCPC Time-Limited Funding
municipalities to replicate
effective programs
Results contribute to
Identify and encourage the development of
Treasury Board
Model and Promising Programs
Reports
Encourage the
Use
development of
Multi-site
Fidelity Tools to
Evaluations
Increase
To Increase
Attribution
Knowledge
5
6. What is Program Fidelity ?
Fidelity in evaluation is used to describe the extent to which the
initiative/intervention corresponds to the originally intended program
The literature uses the following terms interchangeably
Adherence
Compliance
Integrity
6
7. Benefits of a Fidelity Index
A Fidelity Index:
Lists the elements that contribute to the success of the
program
Makes the important elements of the program visible and
helps evaluators “bound” the program
When expected results are unfavourable, model programs can
attribute the results to low fidelity levels while still being able to
maintain a reputation for being an effective program
Identifies what elements practitioners should focus on to
increase compliance
Ensures that elements in the index will have a complementary
quality assurance protocol and definitions for practitioners to
review
7
8. What is a Quantifiable Fidelity Index?
A quantifiable fidelity index is a set of measureable items that
determine whether the program elements have been implemented
as planned.
Each item provides a quantifiable measure using ratio or interval level
data that can be later used to test for statistical significance, changes in
effect sizes or clinical significance
The items in the index can be checked for construct validity. We can
confidently determine if we are measuring what we think we are
measuring. (i.e. Using Chronbach’s alpha and other reliability tests can
be achieved by using quantitative measures)
Composite indices and aggregate scores can be inserted into
multivariate analysis to determine whether the fidelity related scores
contribute to participant related changes.
8
9. How Can a Measureable Fidelity Index Contribute to
Better Program Attribution?
You can confidently answer the question, ‘Did the model program contribute
to change or were additional unintended program elements contributing to
changes ? (i.e. Was the inconsistent implementation of cognitive
behavioural sessions responsible for the lower than expected changes in
anger for boys in the SNAP program?)
Each element that is quantified can be placed into a multivariate causal
model. The evaluator can:
Isolate whether the program’s adherence levels contributed to
favourable or unfavourable change in the outcomes of interest
Isolate what specific aspect of the program contributed to greater effects
or changes in the outcomes of interest
Identify the need to find other explanatory variables that are contributing
to program impact (i.e. R squared adjusted result < .80 informs us that
there are other factors that account for the change in reduction of drugs
or offending for example)
Identify if low fidelity to the program is contributing to variation in the
variables of interest (i.e. Does low participant responsiveness in 25% of
the TND classrooms affect the overall goal of preventing or reducing
drug use in youth?)
9
10. Use of Fidelity in Four NCPC Model Programs
(Based on literature prior to NCPC Implementation)
75% of model programs used 25% of model programs used fidelity
fidelity for for
Process evaluation only Outcome/Impact Evaluations
Method of Attribution is stronger
Method of Attribution is limited to
and can identify how much
Inferences about how
Correlation exists between the
Program is related to outcomes
Program elements and outcomes
10
11. Key Steps in Using a Quantifiable Fidelity Tool to
Increase Attribution
STEP 1: Identify all the key program elements using a comprehensive
framework
STEP 2: Apply appropriate measures
STEP 3: Construct the causal model
STEP 4: Calculate and interpret the coefficients (multivariate
analysis)
STEP 5: Reporting: Enhance program attribution. Determine if
program fidelity levels contributed to outcomes
11
12. STEP 1:
Identify all key elements using a comprehensive framework
Based on Dane & Schneider Framework (1998)
Develop the fidelity tool and include all elements in the following four areas:
1. Implementation: Was the program delivered as intended? (i.e. Were all 8
sessions delivered on a weekly basis over a 2 month period?)
2. Dosage: How much of the program was delivered (i.e. Did the youth receive
the 48 hours of individual case management?)
3. Quality: Are the main components of the program delivered clearly and
correctly?
4. Participant Responsiveness: Does the program stimulate interest among
the participants and practitioners? (i.e. Are teachers interested in
implementing the curriculum to students?)
5. Monitoring Control/Comparison fidelity- What services did the
comparison/control group receive? Did the program adhere to their intention to
provide minimal services or “usual care”?
12
13. STEP 2:
Apply Appropriate Measures
Durlak and Dupre 2008 reviewed 59 studies (1998-2005) that had some
type of fidelity tool and found the following typical measures:
27/59 studies (46%) used categorical measures
Ordinal scales (i.e. assigning a definition to low, moderate and
high fidelity levels)
32/59 studies (54%) used continuous variables/interval measures
Averages reported as percentages (i.e. observations of each
session identified that 65% of the youth showed interest in the
counselling sessions)
Likert scales results were converted to continuous variables
Used actual # of hours, # of sessions, # of months especially for
dosage related measures
13
14. STEP 2:
Sample Treatment Intensity Measures
(Stop Now and Plan® Program )
Sample : Quantifiable fidelity checklist (note: this is not the full checklist)
vTarget Group Population Met: 8/10
v 12 Cognitive Behavioural group Sessions 10/12
Sessions delivered weekly within 3 months
Delivered by a trained facilitator
Approved manuals used (weekly documentation)
v 12 Cognitive Behavioural concurrent parenting sessions 10/
12
Fidelity Level: High T Score: 28/ or 82.4%
otal 34
____________________________________________________________
Fidelity Legend
No Fidelity: 0-29%
Low Fidelity: 30-69%
Moderate Fidelity: 70-79%
High Fidelity: 80-100%
14
15. STEP 3: Construct the causal model
(Multiple Regression Analysis sample)
A multiple regression equation can be used to make estimates
about key program outcomes (i.e. recidivism) based on given
values for a number of explanatory variables (i.e. levels of drug
use, # of hours participating in treatment, and type of
practitioner)
Ensure all statistical assumptions are met before using the model
Construct the causal model by using the following equation :
y = β0 + β1X1 + β2X2 + …. + β kXk + ε
15
16. STEP 4 :
Calculate and interpret coefficients
Interpretations
Overall, we are looking at the marginal contribution that each X
(explanatory) variable is making on the Y (outcome) variable when the
other X variables are being held constant.
Review and interpret the output and consider the following
questions:
Are the elements from the fidelity tool statistically significant?
What elements of the fidelity tool have higher values and contribute
more to change in the key outcomes of interest?
What is the overall contribution of the program elements on the
outcome?
How much does the impact of fidelity have on program outcomes?
16
17. STEP 4:
Calculation and Interpretation of Coefficients
cont’d
B e c le a r a b o u t h o w yo u g o t fro m
yo ur fid e l ity to o l to d isc us si ng
w h e th e r a d h e re n ce le v e ls h a d a n
im p a ct o n p r o gr a m o u tc o m e s …
Th e re a r e n o m ir a cle s h e r e … ..
“I
th i n k y o u s h o u ld
b e m o re e xp lic it h e re
in s te p tw o .”
11
17
18. Step 4:
Calculate and Interpret Coefficients (sample data only)
Y= Aggressive Behaviour
Constant 5.274
Coefficient of (CBS) Cognitive Behavioural Sessions (0-12 hours)
X1 0.4487
Coefficient of (DSH) Duration of Service Hours (total program hours)
X2 0.3334
Coefficient of (FPS) Levels of Fidelity (Participant Responsiveness)
X3 0.2448
R 2 (adj.) 0.7553
18
19. Step 4:
Calculate and Interpret Coefficients
Y= 5.274 - 0.4487 (CBS) - 0.3334 (CPG) - 0.2448 (FPS)
R Square (adj.) =.7553
For every increment of cognitive behavioural sessions, aggressive
behaviour decreases by 0.4487
For every increment of cognitive parental group sessions,
aggressive behaviour decreases by 0.3334
For every increment of fidelity (responsiveness), aggressive
behaviour declines by 0.2448
R square (adjusted) : The treatment intensity of the cognitive
behavioural session, duration of service hours and total group
responsiveness contributed to 76% of the variability in Aggression
levels.
Fidelity levels related to group responsiveness are positively
correlated with the outcomes measuring changes in aggression
levels but contribute less than treatment related factors
19
20. Step 4 Calculation cont’d
Use Multi-Level Analysis
Include fidelity scores in the ANOVA or regression equation
Use a multilevel analysis approach:
Group fidelity measure
Participant responsiveness
Implementation measures
Quality of Implementation
Individual measures
Dosage (quantity)
Ensure all key quantifiable fidelity scores are tested for separate
cohorts (i.e. various classrooms, sites or session). Correlate
fidelity scores with key results (i.e. changes in knowledge,
attitudes and behavioural changes)
20
21. Step 5: Reporting Results
What Really Contributed to the Results?
High Levels of Fidelity Low levels of Fidelity to the
to the Program Program
Favourable Changes Determine whether fidelity scores
or core treatment measures are
Made in the Increased
contributing to results
Outcomes of Interest Confidence that the Qualify results by explaining how
Program is Effective fidelity may have contributed to
(It Works!!) results
Be cautious about reporting on
program attribution
Unfavourable Explore other sources of data to
Changes explain findings:
Increased
Made in the How many elements of the
Confidence that the fidelity index were low?
Outcomes of Interest Program is Ineffective
What elements of the index were
low?
21
22. STEP 5: Determine if Program Outcomes are related to Fidelity
(Towards No Drugs sample)
Outcome Fidelity
Program-specific knowledge 0.33 (0.09)***
Beliefs: health-as-a-value 0.10 (0.04)**
Beliefs: pro-drug myths -0.15 (0.06)**
Cigarette intentions -0.08 (0.04)**
Marijuana intentions -0.09 (0.04)**
Alcohol intentions -0.07 (0.04)*
Multi-level models * p< .10 ** p < .05 *** p < .0001, one
tailed
Rohrbach, L.A., Gunning, G., Sussman, S., & Sun, P. (May; 2008).
Predictors of implementation in the Project Towards No Drug Abuse
dissemination trial.
22
23. Summary
Attempt to utilize 5 categories: Use Implementation, Dosage,
Quality, Participant Responsiveness and Monitoring control group
fidelity categories to ensure the tool/index is comprehensive
Utilize a participatory approach to identify the key elements
Use a statistical approach (backward or forward selection procedures)
Evaluators should advocate for the development and use of a
quantifiable fidelity tool to be implemented in the evaluation
study
Encourage the use of interval levels of measurement to each item in
the tool. This type of measure can provide specific information about
incremental changes in the outcomes of interest
Qualitative information should also be collected to explain why the
results may be favourable or unfavourable
23
24. Summary
Construct a multivariate analysis model that will incorporate
elements of the fidelity tool
Consider using multilevel models that utilize more than one equation
that will take into consideration different levels of data (i.e. Towards
No Drugs fidelity tool required a multilevel model for the analysis:
schools are nested in communities; pupils are nested in schools etc..)
Report how the fidelity levels related to the outcomes of interest
Where fidelity levels and expected results are low, ensure that
triangulation with other data is used to verify program attribution
levels
Explore other elements of the regression or other multivariate
equations used to isolate what program elements may or may not have
contributed to the results.
24
25. Contact Information
Donna Smith-Moncrieffe, BSc., Crim Dip, MSc.
Senior Evaluation Advisor
Public Safety Canada
National Crime Prevention Center
Policy Research and Evaluation Division
E-mail: donna.smith-moncrieffe@ps.gc.ca
25