Gabrielle Wills
18 August 2015
The author is grateful for the assistance given by Dr Martin Gustafsson at the Department of Basic Education in understanding the data used
for this study, providing insights into the policy context and comments on draft papers of this work. Comments from Prof.Servaas van der Berg
and other ReSEP colleagues are also acknowledged.
Introduction & background
• There is convincing evidence that school principals matter
considerably for student learning across both education &
economics literature:
o Leithwood et al (2004) – review of case studies of school
leadership in the U.S. and Europe: school principals are only 2nd
to teachers in terms of their importance for student learning &
school effectiveness in general.
“Indeed, there are virtually no documented instances of troubled
schools being turned around without intervention by a powerful
leader. Many other factors may contribute to such turnarounds, but
leadership is the catalyst” (Leithwood et al., 2004: 7)
Introduction & background
• Growing evidence in the U.S. from value-added models using large-scale
datasets that school principals matter a lot for student learning.
Valueadded
models
• Identify the additional value that principals bring to learning
after isolating out the contributions of individual teachers, the
school & the ability & backgrounds of individual students.
 A highly effective principal can raise the achievement of the average student by
2-7 months of learning in a year (Branch, Hanushek & Rivkin 2012; see also Coelli & Green
2012; Grissom, Kalogrides & Loeb 2012; Chiang, Lipscomb & Gill 2012).
o Only 2nd to the direct effects of individual teacher quality on learning but
principals affect all students in a school.
o The overall impact from increasing principal quality substantially exceeds
the benefit from a comparable increase in teacher quality.
What about policy?
“Principals form part of a strategic sector that has not been duly
explored in its potential for contributing to education progress.”
-
Weinstein, Munoz & Raczynski 2011:298
 In SA, there have been notable shifts in raising the value of school leadership
& management as critical levels for learning gains.
- Amendments to legislation, statements & action of the DBE & in national
policy plans
 Education Amendment Act 2007
- Increased accountability for principals: requirement to plan for
academic improvements and report against targets set.
o But does this induce the kind of behavioural change that results in
more learning?
o Form vs. substance- Taylor 2014; “Buffering”-Elmore 2000
- Allows for temporary replacement of principals in
underperforming schools.
Policy progress
 Standards for Principalship – draft policy
- Redefines the roles and responsibilities of school principals
- More focus on instructional leadership than PAMS job
description
 IQMS
- Implementation challenges. CMs not evaluating principals in
many circuits; too easy to get high ratings (DBE, 2012: 44**)
- Are ‘carrots & sticks’ effective to induce the right kind of
behavioural change?
- ELRC Collective Agreement No. 1 2008: Promising performance
mgt. system for principals and deputies but withdrawn at ELRC
negotiations.
**DBE (2012) IQMS Annual Report 2011/12, Pretoria: Department of Basic
Education.
National Development Plan
The need to strengthen the policy framework governing principals has
arguably gained traction through the NDP which proposes policy
improvements in three areas:
A) Improve the principal
appointment process
B) Managing their
performance
• Competency based-testing
• WC and GP already
doing this.
• Increase min.
qualifications to include
having an ACE in school
management and
leadership
• Performance contracts for
school principals
• Replace underperforming
principals with better ones
See: NPC (2012), pp 309-310
C) Greater powers over
school management
• More autonomy for
principals in functional
schools
• But hiring and firing
remains at the provincial
level
SA Literature
 Hoadley and Ward (2009) in their review of literature on
school management and leadership reiterate earlier
remarks by Bush et al (2006) …
Our understanding is limited of how the actions and
behaviours of school leaders in South Africa are
contributing to or detracting from school functionality,
particularly with respect to producing learning
outcomes.
But take a step back….
Research Objectives
But there has been a lack of empirical evidence in SA to guide
& support policy developments in this area.
Objective 1)
• Identify the overarching quantitative characteristics of the
labour market for principals in SA.
Objective 2)
• Using this evidence to inform, support and debate recent
policy developments.
• Consider the relevance of the NDP policies in light of the
evidence and propose additional policy recommendations.
Research Questions
 Who has been appointed to assume responsibility for leading





schools?
- Demographics
- What are their qualification levels and years of experience?
Do principal characteristics systemically differ across poorer and
wealthier parts of the schooling system?
What are the mechanisms driving these differences in how
principals are distributed across schools?
How much churning is there in terms of both in terms of
attrition related moves and within system transfers?
Do incentives exist in the system that direct the transfer of
principals across schools in ways that aggravate existing
inequalities in the distribution of principals?
Do credentials, as measured in terms of qualifications and
experience, provide a signal of principal quality in SA?
Data
EMIS
Persal
Sept 2004, Oct 2008,
Masterlist of schools
79-89% matched
Oct 2010, Nov 2012
(6-10% of non-matching due to vacancies)
Constructed panel
dataset with 4 “waves”
Annual Survey of Schools
(some)
SNAP over time
Matric Examination Data
series
(Gustafsson & Taylor, 2012)
1) The aging profile of school principals
Principal replacements: substantial & increasing
Age distribution of principals, 2004 & 2012
30
Percentage
25
26 27
26
2004: Average age 48 yrs
2012: Average age 51 yrs
25
23
20
19
2004: 17% aged 55+ yrs
2012: 33% aged 55+ yrs
15
14
13
10
10
8
5
0
2 0
26-34
4
3
35-39
Source: Wills (forthcoming)
40-44
45-49
50-54
Age of principals
2004
2012
55-59
60+
> 7000 principal
replacements (20122017)
No major differences in the age profile by phase level
Age profile of principals in 2012 by the phase level of the school they lead
35
30
30
Percentage
25
20
26 27
2.6 X
Replacements:
22
primary schools16
14
15
24 24
26
23 23
Replacement
secondary
schools
14
8
10
6 7
3 4 3
5
0.2 0.3 0.1
0
26-34
35-39
40-44
45-49
50-54
Age of school principals
Primary/Intermediate
Combined
55-59
Secondary
60+
An aging profile of school principals
School principals in 2012 aged 55 years or older, by quintile
2500
48.5
50
2000
40
30
36.2
27.2
40.1
1500
28.9
1000
20
500
10
0
0
Quintile1
Quintile2
% of principals aged 55 years or older
Source: Wills (forthcoming)
Quintile3
Quintile4
Quintile5
Number of principals aged 55 years or older
Number of principals
Percentage of principals
60
An aging profile of school principals
"The imminent retirement of the majority of principals brings both
challenges and new opportunities for OECD education systems. While
it means a major loss of experience, it also provides an
unprecedented opportunity to recruit and develop a new
generation of school leaders with the knowledge, skills and
disposition best suited to meet the current and future needs of
education systems“ (Pont, Nusche and Moorman, 2008: 29)
 Policies need to be crafted to ensure that the right leaders are
positioned in schools.
- Although too slow, some of this opportunity still can be leveraged.
2) The unequal distribution of school
principals
The unequal distribution of principals
Principal qualifications (REQV), 2012
100%
90%
16
21
18
14
REQV 13
29
REQV 14
31
34
60%
38
34
32
50%
33
40%
30%
28
REQV 16-17
27
20%
21
14
19
24
31
38
0%
All schools
REQV 15
30
29
10%
3
28
80%
70%
8
Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5
Source: Persal-EMIS matched dataset, own calculations. Notes: …Percentages add up to
100 per cent in each sub-group.
The unequal distribution of principals
 Unique to SA
- Sorting of principals to schools in accordance with
institutional Apartheid policies.
o Average principal in 2012 has 25 years of service.
Positioned into schools as teacher during Apartheid.
o Over 55% of principals in South Africa are promoted
from within the same school.
 But patterns of sorting have persisted…
Unequal distribution of principals
Educational qualification of outgoing & newly appointed principals, 2010 & 2012
100%
Quintile 2
Quintile 3
REQV 15
Quintile 4
39
REQV 16-17
Incoming
29
Outgoing
22
Incoming
Incoming
Quintile 1
12
21
Outgoing
Outgoing
All schools
14
15
Incoming
8
20
Outgoing
12
27
21
Incoming
15
Incoming
Outgoing
0%
26
REQV 14
26
29
21
17
36
31
23
REQV 13
31
33
20%
10%
26
35
40%
29
27
32
32
50%
35
REQV 10-12
31
41
46
8
33
37
40
5
18
24
26
33
60%
30%
16
25
80%
70%
16
10
12
Outgoing
90%
14
Quintile 5
Source: Persal-EMIS matched dataset. Notes: …Percentages add up to 100 per cent in each sub-group.
Unequal distribution of principals
Figure 1: Average years of service of incumbent, outgoing and new principals, 2012
35
Years of service
30
29
25
25
20
20
29
28
25
24
18
32
30
31
26
26
27
22
21
20
21
9
9
15
10
10
10
All schools
Quintile 1
11
9
5
0
Quintile 2
Quintile 3
Quintile 4
Quintile 5
Outgoing principal (2008 or 2010)
Incumbent principals (2012)
Incoming principal (2010 or 2012)
Estimated total years as principal
Source: PERSAL 2008, 2010 and 2012. Notes: Incumbent principals are identified as the senior principal in a school in
2012. Outgoing principals are senior principals who leave the public education system either between 2008 and 2010 or
between 2010 and 2012. Incoming principals are senior principals in either 2010 or 2012 who were not identified as
principals in PERSAL in previous periods. Years of service in public education are not necessarily equivalent to total years
of experience in teaching/school leadership if principals had worked outside of the public education sector. However it is
likely to provide a very close proxy.
Two sources underlying principal sorting
Inequity in
principal sorting
across schools
1) Preferences of
principals
a larger pool of well-qualified candidates
applying for posts in wealthier schools?
2) Variations in the
recruitment and
selection process
across schools
Wealthier schools impose more
stringent appointment criteria?
3) Traditional qualifications as captured
on payroll do not signal principal
quality
Credentials as a signal of quality?
 Education systems internationally reward teachers on the
basis of qualifications and seniority.
 Qualifications and seniority to some extent guide
promotions (RSA DoE – PAMS 2003).
 Few rewards for performance other than IQMS 1% salary
progression.
 But do credentials i.t.o traditional qualifications &
experience actually signal quality?
- Little international evidence to support this
(Clotfelter, Ladd & Vigdor, 2010; Hanushek, 1986; Hanushek 2007,
Ballou & Podursky 1995; Clark, Martorell & Rockoff 2009).
Credentials as a signal of quality?
 Various challenges associated with estimating reliably how
principal credentials are related to school performance.
- Different types of principals are attracted to different types
of schools.
- Certain principals may attract or be attracted to different
types of students.
- Using the panel dataset constructed for the study, an
attempt was made to control for some of these patterns of
sorting to schools that may bias estimates.
 Do not need to control for principal quality given the research
question: the ability bias forms part of the effect of interest
(Clark et al 2009).
Credentials as a signal of quality?
 A subset of the matched payroll-EMIS dataset was used for
the estimation:
• schools that had gr. 12s in 2008, 2010 &
2012
• connected to school matriculation
examination outcomes in these years
(Gustafsson & Taylor, 2013)
Average mathematics % among mathematics takers
OLS 1
OLS 2
OLS 3
FE 1
FE 2
FE 3
REQV level
1.704***
0.216**
0.213**
0.739***
0.637***
0.573**
(continuous)
(0.116)
(0.091)
(0.091)
(0.252)
(0.245)
(0.240)
0.001
-0.035*
-0.041**
-0.033
-0.059
-0.085**
(0.029)
(0.020)
(0.020)
(0.042)
(0.041)
(0.041)
X
X
X
X
X
X
X
X
X
X
Years of service
Principal controls
School controls
Year fixed effects
R-squared
X
0.045
0.457
0.46
Within R-squared
N
13139
13093
13093
N (clusters)
F
63.513
301.032
X
277.956
0.008
0.084
0.096
13139
13093
13093
4460
4460
4460
6.215
46.93
49.444
Source: Matched Persal-EMIS dataset, with matriculation data. Notes: The unit of observation is school-year.
REQV is entered as a continuous variable from ranging from 10 to 17. Time-varying principal controls include
gender, age and their position in the previous data period observed. Time-varying school controls include total
school enrolment and its square, number of teachers per 100 students and the percentage of students that are black.
Additional time invariant school controls are included in the OLS regressions. These are school quintile, former
department classification, urban location and provincial dummies. Year dummies are entered for 2010 and 2012.
The year reference category is 2008. Statistically significant at * p<0.1, **p<0.05, ***p<0.01. Standard errors are
% of examination takers who achieve the National Senior Certificate
OLS 1
OLS 2
OLS 3
FE 1
FE 2
FE 3
REQV level
2.798***
0.456***
0.445***
1.546**
1.506***
0.533
(continuous)
(0.203)
(0.173)
(0.167)
(0.619)
(0.579)
(0.468)
0.013
-0.075**
-0.121***
0.371***
0.244***
-0.112*
(0.045)
(0.034)
(0.033)
(0.085)
(0.079)
(0.065)
X
X
X
X
X
X
X
X
X
X
Years of service
Principal controls
School controls
Year fixed effects
R-squared
X
0.043
0.37
0.408
Within R-squared
N
13 490
13 442
13 442
N (clusters)
F (p-value)
68.838
439.008
X
409.027
0.04
0.209
0.294
13 490
13 442
13 442
4 503
4 503
4 503
23.955
106.09
187.188
Source: Matched Persal-EMIS dataset, with matriculation data. Notes: The unit of observation is school-year.
REQV is entered as a continuous variable from ranging from 10 to 17. Time-varying principal controls include
gender, age and their position in the previous data period observed. Time-varying school controls include total
school enrolment and its square, number of teachers per 100 students and the percentage of students that are black.
Additional time invariant school controls are included in the OLS regressions. These are school quintile, former
department classification, urban location and provincial dummies. Year dummies are entered for 2010 and 2012.
The year reference category is 2008. Statistically significant at * p<0.1, **p<0.05, ***p<0.01. Standard errors are
Credentials as a signal of quality?
 Little evidence that traditional qualifications as captured
on payroll are signalling quality.
- This does not imply that principals don’t add value.
Rather this value is not captured in traditional credentials
and particularly the REQV system.
- But principals are upskilling on the job. Higher pay for
higher REQVs but not necessarily bringing more value 
Rent extraction.
4) Low levels of churning
Low levels of churning
Turnover rates for principals
Average yearly turnover rates
Turnover over the period
(lower bound) over the period
2004-2008
23.4
5.8
2008-2012
28.7
7.2
2008-2010
13.6
6.8
2010-2012
16.6
8.3
Source: Persal-EMIS matched data Notes:… ^Other educators include teachers, departmental heads and deputy
principals who can be matched to an ordinary school in EMIS data.
• Low levels of principal turnover compared with other local
benchmarks on employee turnover the public sector & then the
U.S. (Miller 2013, Beteille et al 2012)
2004-2008
77 %
“Stay put”
23 %
Turnover
2008-2012
Attrition:
67%
71%
“Stay put”
29 %
Turnover
Attrition:
75%
Low levels of churning
 Low levels of mobility implies high tenure.
 Two added features of low levels of churning:
- Over 55% of principals are promoted from within schools
- Less than 3% of moves are across provincial moves
 Various reasons for low mobility incl. lack of accountability measures
affecting principals.
- In the U.S. job security concerns incentivize principals to move
from weaker performing to better performing schools.
- Despite IQMS, less than 1 in 1000 principals were dismissed in SA
in 2011.
- BUT there are other incentives at play that affect mobility patterns
e.g. larger schools for higher pay.
Summary
Rising age profile of principals implies a substantial & increasing no. of
principal replacements in schools.
- proportionally more retirements are taking place in wealthier schools, but
higher absolute demand for principal replacements in the poorest schools.
2. Principals are unequally distributed across schools with less qualified & less
experienced principals overly represented in poorer schools.
- Initial matching of new principals to schools continues to persist in line with
historical patterns.
3. The value principals bring to schools is not signalled through their observed
credentials as captured in payroll: REQV levels and years of service.
4. Despite rising levels of retirement related attrition, low levels of mobility &
high levels of average tenure characterise this market. While the no. of within
sector transfers is low transfer patterns tend to operate in the same direction
as existing inequalities.
1.
Policy recommendations
Broad recommendation 1)
• Policies should be aimed at improving the initial match of
principals to schools while developing incumbent principals over
the length of their tenure.
Broad recommendation 2)
• Where qualifications and experience provide weak signals of
quality, policies guiding the selection or principals and those that
reward performance should extend beyond observed credentials
to identify expertise and skills that may be better quality signals.
A) Improve the principal appointment process
Relevancy
Competency based-testing
• HIGH
• Substantial & increasing no. of
principal replacements
• Target the initial sorting of
principal into schools to
address unequal distribution
• Traditional academic
qualifications are poor signals
of principal quality. Other
competencies are likely to
matter more.
Increase min. qualifications to having an ACE
in school management &leadership
• LOW TO MEDIUM
• Will this reduce the available
pool of candidates?
B) Performance Management?
Relevancy
Performance contracts for school
principals
• HIGH
• With low levels of principal
mobility, it is necessary to
improve the calibre of
incumbent principals over the
course of their tenure.
• Reward performance rather
than qualifications.
• But careful thought to design
and implementation!
Replace underperforming principals with
better ones
C) Providing principals with greater power?
Relevancy
Providing principals with greater powers over school management
• No local evidence linking management powers to increased
learning in schools.
• International evidence generally supports the decentralisation of
decision making to the school-level in raising school outcomes
(Hanushek & Woesmann 2007).
• However increased autonomy must be packaged with
accountability measures
• “Local autonomy without strong accountability may be worse
than doing nothing” -Hanushek & Woesmann, 2007:74
Conclusion
 The urgency to implement policy to support the right
appointments of new principals cannot be reiterated
enough in light of the substantial & increasing number of
principal replacements to be made in schools.
 With each principal placement, the leadership trajectory of
the average school is established for almost a decade.
 Evidence-based policy making has a strong role to play in
getting this right.
Additional slides
Gender imbalance in school leadership
Figure 2: Percentage of principals and teachers that are women by school phase level
100
90
79
80
76
71
Percentage
70
57
60
50
46
40
43
30
36
29
27
20
10
19
34
15
0
Primary/
intermediate
Combined
Secondary
All schools
Source: Persal-EMIS matched dataset. Notes: Teachers include
deputy principals and heads of department.
Principals (year = 2012)
Principals (year = 2004)
Teachers (year = 2012)
Incentives: Moves to urban & larger schools
Multinomial logistic regression predicting the probability of
1) Moving vs staying and 2) Leaving vs. staying:
 Principals are less likely to move to another position in the
system if they are initially in an urban school post rather
than a rural school post.
 School size also influences transfer patterns. The predicted
probability of principal turnover is inversely related to
school size.
 Secondary school principals are considerably more likely to
leave the system and to transfer to another position within
the system than principals in primary or intermediate
schools
Incentives: Moves to wealthier schools?
Quintile of sending school
Quintile of receiving school
1
2
3
4
5
Total
221
107
82
24
4
438
50.5%
24.4%
18.7%
5.5%
0.9%
100%
71
99
61
12
12
255
27.8%
38.8%
23.9%
4.7%
4.7%
100%
49
56
100
22
5
232
21.1%
24.1%
43.1%
9.5%
2.2%
100%
7
16
26
34
22
105
6.7%
15.2%
24.8%
32.4%
21.0%
100%
7
8
14
8
56
93
7.5%
8.6%
15.1%
8.6%
60.2%
100%
355
286
283
100
99
1 123
31.6%
25.5%
25.2%
8.9%
8.8%
100%
1
2
3
4
5
Total
Source: Persal-EMIS matched dataset. Notes: Transition matrix is calculated for school principals in 2008 (or 2010) who move to another principal post in different
school by 2010 (or 2012). For this group, 1 158 transitions should be observed but data is missing on quintile ranking for some schools. Frequencies are in the top of each
cell, and percentages are at the bottom. Quintile rankings refer to DBE rankings.
Incentives: Students’ vs. principal’s race
Probability of moving by principals’ race and the student race composition
(Multinomial logistic regression predicted probabilities)
The race of the
principal relative to
the race of the
student composition
of the school
appears to be
associated with
patterns of transfer
in the first 20042008.
Are the most qualified more likely to leave the
system?
The probability of turnover by the REQV qualifications of principals
(Multinomial logistic regression predicted probabilities)
On the contrary:
The most qualified
principals are less
likely to move out of
the system than less
qualified, but more
likely to move
within the system.