E (p) owering Your Institution A Mixed-Model Approach to Assessment © 2004 Douglas Joubert Douglas James Joubert Greenblatt Library
Nov 07, 2014
E(p)owering Your InstitutionA Mixed-Model Approach to
Assessment
© 2004 Douglas Joubert
Douglas James Joubert
Greenblatt Library
Objectives
• Creating a culture of assessment
• H1 and H3
– Data Considerations
– Sample Characteristics
– Statistical Analysis
• Collaboration & Innovation
Illusions of Certainty1
Retaining and growing their customer base, and focusing more energy on meeting their customers’ expectations is the only way for academic libraries to survive in this volatile environment
--Rowena Cullen, 2001
Creating a Culture of Assessment
Culture of Assessment
Benchmarking
ARL Service Academy
AAHSL Task Force Questions
LibQUAL+
“Aspirational” “Comparator”
Speaking a common languageQuantitative
Input
Influential Players
Input
Mixed Model Approaches
Linear Model of Research
Theory
H1
Sampling
Analysis
Validation
Adapted from Flick, 2002
Circular Model of Research
Adapted from Glaser & Strauss, in Flick, 2002
Comparing
ComparingComparing
Sampling Sampling
CaseTextual analysis
CaseTextual analysis
CaseTextual analysis
PA T
Quantitative Research
ToMaze
Quick! get to the
cheese
Qualitative Research
I really don’t like cheese
I need a vacation
Do these frames
make me look smart?
ToMaze
Why am I here?
Qualitative Research
“Triangulation”
“Grounded Theory”“Bricoleur”
TwoVariables
?
Bothcontinuous
?
Prediction?
r/s between 2variables?
Both variables normal?
Pearsoncorrelation
Rankcorrelation
Linear regression
Y Y
N
Y
Y
N
Y
Statistical Inference Flowchart
Adapted from Rosner, 2000
Overall Data Considerations
24th AAHSL 25th AAHSL
Local SPSS Tables Local SPSS Tables Local SPSS Tables
2002 LibQUAL+
Overall Data Considerations
24th AAHSL 25th AAHSL
Local SPSS Tables Local SPSS Tables Local SPSS Tables
N > 13,000 2002 LibQUAL+
Overall Data Considerations
24th AAHSL 25th AAHSL
Local SPSS Tables Local SPSS Tables Local SPSS Tables
Descriptive
2002 LibQUAL+
Overall Data Considerations
• Around 120 cases• Formatted for
reading, not for statistical analysis
• Formatting must be monitored
• Are missing scores “systematic” or “random”?
AAHSL Data
Overall Data Considerations
• Over 13,000 cases• In SPSS, built for
statistical analysis• Formatting and
Missing scores built into the design of the variables
LibQUAL+ Data
Overall Data Considerations
• Use independent reviewer before data migration• Determine pattern of missing data before migration• How are you going to deal with missing scores• Easier to spot errors with 120 cases than with 13,000
AAHSL Data
LibQUAL Data
The First AAHSL Group
1. How do satisfaction ratings differ for various reporting structures?
2. How does the size of library staff affect satisfaction ratings?
3. How does the number of constituents affect satisfaction ratings?
4. How does the ratio of staff to constituents affect satisfaction ratings?
Source: AAHSL Task Force on Quality Assessment
Hypothesis 1
Hypothesis I (H1) was operationalized as, “What is the affect of library reporting structure on the mean
measure of overall quality of service?”
H1 Data Considerations
• Not originally included in the study design• Included to rectify a priori considerations for
the remaining three hypotheses
“person-levelsubscales”
“institution-levelsubscales”
means by dimension for each person
means by dimension for each institution
H1 Data Considerations
• Medical School• Other Health Science
School• University Library• Health Science Center
Administration• University
Administration• Other
• Question 3 (Section 5.6) of the 2002 LibQUAL
Library Reporting Structure
Overall Quality of Service+
Defined by the 2001-2002 AAHSL Library Statistics
Survey
H1 Sample Characteristics
• 35 AAHSL institutions contributed data for the 24th Annual Survey and participated in 2002 LibQUAL+.
Health Science (14) University Administration (2)
Medical School (12) Other (2)
University Library (3) Missing (2)
H1 Unit of Analysis
• This analysis needed to demonstrate if OVERSAT1 differed by reporting structure.
• H01 required an analytical method for dealing with the institutional differences
Variable reserved from LibQUAL data
Library reporting structure
Differences not explained by the grouping
Differences explained by the grouping +
H1 Unit of Analysis
• A Linear Mixed Model (LMM) allowed me to control for fixed and random effects
OVERSAT [DV] = LIBSTRUC [IV1] + INSTID(LIBSTRUC) [IV2]
Fixed-effectsRandom-effects
Descriptive Statistics for Overall Quality of Service
How would you rate overall quality of the service provided by the library?
Library Reports To Count Mean SD CV (%)
Medical School 4868 7.23 1.451 20.1%
University Administration 879 7.18 1.583 22.0%
Health Science Center 5972 7.50 1.323 17.6%
Other 649 7.22 1.457 20.2%
University Library 759 7.19 1.393 19.4%
Total 13127 7.35 1.408 19.2%
Type III Tests of Fixed Effects for Overall Quality of Servicea
Source Num df Den df F Sig.
Intercept 1 26.927 6640.536 .000
LIBSTRUC 4 27.038 .924 .464
a. Dependent Variable: How rate overall quality of the service?
Hypothesis 3
Hypothesis 3 (H03) was operationalized as, “How does the number of constituents served affect
satisfaction ratings?”
H3 Data Considerations
Secondary Clientele Served
Table Q4
Academic and Primary Hospital Staff (STAFFTOT) = NO
Primary Clientele Served
Table Q3A-Q3C
Total constituents served (CONSTOT)
1. Faculty (FACTOT),
2. Interns, residents, and fellows (IRF)
3. Students (STUTOT)
H3 Data Considerations
• From Question 3 (Section 5.6) of the 2002 LibQUAL computed mean overall quality of service by institution
“person-levelsubscales”
“institution-levelsubscales”
Mean OQS for each person
Mean OQS for each institution
H3 Unit of Analysis
• Because of the findings from H1, H3 could be analyzed with correlation and simple linear regression models
CONSTOT M_OQS by Inst
IV DV
Descriptive Statistics for H3
Dealing with Outliers• Distinct score relative to the bulk of
the distribution• Not all outliers are influential• Cook’s D is an index based on the
F-statistic• Cook’s D > 1 deemed an outlier• Cook’s D = 3.10679
X-DirectionIV
Dealing with Outliers• Studentized (SRESID) residual provides
information about outliers in the “Y” direction
• Framed within the context of the SND, so scores > ± 3
• Sign is irrelevant because we are only interested in how far the point is in the tail of the distribution
• SDRESID =1.75842
Y
Direction
DV
Dealing with Outliers
• Removing outliers in an objective method is ethical
• Filter variable ($_filter) created to exclude score 4 from the analysis
• Correlation and linear regression was performed with (non-transformed) and without (transformed) the data from case 4
DV
IV
32N
.027Sig. (2-tailed)
-.391*Pearson Correlation
Total Constituents
Served
LibQUAL+ Mean Overall Quality of Service by Institution
* Correlation is significant at the 0.05 level (2-tailed).
Transformed Correlations
Total Constituents as a Predictor of Measures of Overall Quality of
Service
.153.3911,305.423-0.3910.355-8.10
R2RdfFStd. ErrB
• Predictors: (Constant), Total Constituents Served• Dependent Variables: Mean Overall Quality by Institution
Total Constituents as a Predictor of Measures of Overall Quality of
Service
Total Constituents accounts for 15.3% of the variance of M_OQS
X Y .153.391
R2R
Further Investigation
• Controversy on Significance Testing– Evaluating Results Replicability– Confidence Intervals for Effect Size– Emphasize Effect-size Interpretation– β versus Structured Coefficient exploration
for effect size
Thompson, 2002, 1996, and 1992
BuildingConnections
BiostatisticsBiostatistics
AAHSLAAHSL
Peer NetworksPeer Networks
NewDirections
Qualitative
2005 LibQUAL
Local Projects
Collaboration & Innovation