ARE TEACHER ABSENCES WORTH - IA de Investigación especializada en el MIT

ARE TEACHER ABSENCES WORTH

WORRYING ABOUT IN THE UNITED

STATES?

Charles T. Clotfelter

(Autor correspondiente)

Departamento de Economía

Universidad de Duke

Box 90245

Durham, CAROLINA DEL NORTE 27708-0245

charles.clotfelter@duke.edu

Helen F. muchacho

Departamento de Economía

Universidad de Duke

Durham, CAROLINA DEL NORTE 27708-0245

hladd@duke.edu

Jacob L. Vigdor

Departamento de Economía

Universidad de Duke

Durham, CAROLINA DEL NORTE 27708-0245

jacob.vigdor@duke.edu

Abstracto
Using detailed data from North Carolina, we examine
the frequency, incidence, and consequences of teacher
absences in public schools as well as the impact of a
policy designed to reduce absences. The incidence of
teacher absences is regressive: when schools are ranked
by the fraction of students receiving free or reduced price
lunches, teachers in the lowest income quartile average
almost one extra sick day per school year than teachers
in the highest income quartile, and schools with persis-
tently high rates of teacher absence were much more
likely to serve low-income than high-income students.
In regression models incorporating teacher ﬁxed effects,
absences are associated with lower student achievement
in elementary grades. Finalmente, we present evidence that
the demand for discretionary absences is price elastic.
Our estimates suggest that a policy intervention that
simultaneously raises teacher base salaries and broad-
ens ﬁnancial penalties for absences could both raise
teachers’ expected incomes and lower districts’ expected
costos.

C(cid:2) 2009 American Education Finance Association

115

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

INTRODUCCIÓN

1.
Whatever the importance of strong training, classroom experience, or ad-
vanced pedagogical methods for the scholastic development of students, estos
factors can have scant effect on a day when a teacher is absent from school.
Teacher absences are an endemic problem in developing countries (Banerjee
and Duﬂo 2006; Chaudhury et al. 2006). Baseline teacher absence rates in the
range of 20 a 44 percent have been reported in studies of policy interventions
in Kenya and India (Glewwe, Ilias, and Kremer 2003; Duﬂo and Hanna 2005).
Interventions designed to reduce teacher absence, or improve teacher perfor-
mance generally, have met with mixed success in these settings (Banerjee and
Duﬂo 2006).

The rate of teacher absence in the United States is much smaller than
in these developing countries, and the availability of substitute teachers may
further lessen the potential harm from teacher absences in this country. Pre-
vious studies suggest absence rates for teachers in the United States on the
order of 5 por ciento, or about 9 days per 180-day working year.1 Perhaps for
this reason, there is surprisingly little research on teacher absences in the
United States. Compared with workers in other occupations, sin embargo, amer-
ican schoolteachers appear to have relatively high rates of absence. By com-
parison, ostensibly similarly measured rates of absenteeism due to sickness
average less than 3 percent in the U.S. workforce as a whole.2 This introduces
the possibility that policies speciﬁc to public education have contributed to
the elevated absence rate and that other policies could be used to reduce it.3
Potential social gains from reduced absenteeism include improved student
discipline and achievement and reduced expenditures on substitute teachers.4
Previous literature provides conﬂicting evidence on whether teacher absences
are consequential for student achievement in America, where certiﬁed sub-
stitute teachers are widespread (Ehrenberg et al. 1991; Molinero, Murnane, y

Ehrenberg et al. (1991), who conducted a survey of 381 school districts in New York State in the mid-
1980s, found that teachers took an average of 8.9 days of leave a year. Podgursky (2003) cites a study
of New York City schools in 2000–1 showing an average of 11.3 days a year and a U.S. Departamento
of Education survey concluding that 5.2 percent of teachers were absent on any given day. Focusing
only on sick leave, Bradley, Verde, and Leeves (2005, mesa 1) report rates for Queensland, Australia,
of about 3 percent and a similar rate based on another study in the United Kingdom.

2. Measured as a percentage of hours missed due to illness, maternity or paternity leave, or child care
or other family obligations, the rates of absence in 2005 eran 2.3 percent in the public sector and
1.7 percent in the private sector. In two similar occupations, fue 2.4 percent in community and
social services and 2.7 percent in health care support (A NOSOTROS. Bureau of Labor Statistics 2006).
In addition to the generosity of leave policies, explanations given for the higher absenteeism of
teachers include the high rate of infectious illnesses carried by students, the stress of the job, y
the expectation that teachers will stay out of school to care for their own children. Such expectations
and leave policies are consistent with the notion that teaching as an occupation has traditionally
been made to suit working mothers, with short workdays and summers off to accommodate the
demands of child rearing (Podgursky 2003).

4. One estimate of the cost of substitutes due to excessive teacher absences is on the order of 0.5

percent of total per pupil expenditures (Roza 2007, pag. 5).

116

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Willett 2007). Absenteeism may also have a regressive impact, in which case
interventions to reduce it could promote equity as well as efﬁciency.5

This article aims to address the questions of frequency, incidence, y
effect as well as the potential impact of leave policy, using data on public
schools in North Carolina. We show that the pattern of absence across schools
in North Carolina has a disproportionate impact on low-income students.
When schools are ranked by the fraction of students receiving free or reduced
price lunch, teachers in the lowest quartile average almost one extra sick day per
school year compared with teachers in the highest quartile. We also document
that elementary students perform worse on standardized tests when they are
assigned to teachers who take more absences. This relationship persists in
models that incorporate teacher ﬁxed effects. This ﬁnding corroborates the
work of Miller, Murnane, and Willett (2007). The estimated magnitude of the
achievement effects is small, but aggregated across all students in a classroom
they imply a non-negligible impact of absences on aggregate achievement.

Our study of the potential impact of leave policy on absences follows several
existing studies on the subject. Ehrenberg et al.’s (1991) detailed analysis of
teacher contracts for a large sample of New York school districts revealed
that certain provisions were associated with higher usage of leave.6 Currently
some districts in the United States offer bonuses for teachers who take a
minimal number of absences; one district rafﬂed off a new car among all
teachers with perfect attendance. At the school level, some schools distribute
to teachers at the end of the year any unused funds earmarked for substitute
teachers.7 A randomized evaluation of a comparable bonus program for teacher
attendance in India found that the teacher absence rate was halved in treatment
escuelas (Duﬂo and Hanna 2005). Además, administrative rules covering
teachers’ reporting of absences have also been associated with differential rates
of teacher absenteeism (Imants and Van Zoelen 1995).8

5. One of the rare stories in the general readership news media that did touch on teacher absences,
published in the Chicago Tribune, illustrates some of the issues lurking beneath the surface. Basado
on analysis of several years of data for the Chicago public schools, the Tribune reported chronic
absenteeism concentrated among some teachers in a subset of the district’s schools. In twenty-
two elementary schools, most of which served poor and minority students, per teacher absences
averaged more than twenty days a year—a rate lower than that observed in developing countries but
quite substantial in a local context. In these schools, substitutes attempted to do little teaching, y
discipline deteriorated (Dell’Angela and Little 2006).

6. These provisions included larger than average numbers of leave days allowed overall, larger numbers
of days for bereavement leave, the presence of “sick leave banks,” whereby teachers may borrow
leave days not used by other teachers, and smaller numbers of contractually speciﬁed professional
leave. The authors also found that policies specifying the “buyback” of unused sick days—in cash
or in the form of additional retirement beneﬁts—appeared to inﬂuence the use of sick leave, con
more generous buyback rates associated with fewer absences.
For examples of some current policies regarding teacher absences, see Hobbs 2002, Kossen 2006,
y graham 2006.
For a number of references to studies of practices designed to reduce teacher absences, see also
Molinero, Murnane, and Willett (2007).

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

117

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

We evaluate the impact of a North Carolina policy that permits teachers
to continue taking sick days once they have exhausted their supply of “free”
días, at the cost of $50 per day. The dependence of available sick days on the duration of a teacher’s employment history and the number of sick days taken in prior years generates idiosyncratic variation in the point at which teachers face a price increase. Our estimates, derived from a modiﬁed form of survival analysis, indicate that the likelihood of taking an additional sick day, on the margin, is signiﬁcantly lower when the cost is $50 rather than zero. Back-of-
the-envelope calculations suggest that applying the $50 penalty to all sick days starting with the ﬁrst would reduce the average number of sick days taken by slightly more than one, or about 15 por ciento. Given the savings that would result from cutting back on the use of substitute teachers, districts could raise base salaries sufﬁciently to increase teachers’ expected income while still realizing cost savings. The second section describes the data used in the present study and sum- marizes the broad patterns; the third analyzes correlates of absenteeism; the fourth focuses on the distribution of teacher absence across schools; the ﬁfth examines the effect of teacher absence on student achievement; and the sixth evaluates the impact of ﬁnancial penalties on absence taking. The article ends with a brief conclusion. 2. DATA Our data on teacher absences were provided by the North Carolina Department of Public Instruction, the central state agency that collects uniform adminis- trative data from all of the state’s school districts.9 These administrative data cover the years 1994–95 to 2003–4. We were able to link them to other ad- ministrative records on teachers by means of identifying numbers, supplied by the North Carolina Education Research Data Center, encrypted to preserve conﬁdentiality. For virtually all teachers, the data set gives the number of days absent, by pay period and reason, and for most purposes we aggregate these data into annual totals for each teacher by year. As is common in public school districts across the country, teachers in North Carolina are permitted to take a limited number of days off from work for sickness without losing pay or beneﬁts, and these may include days when a child is sick or a doctor visit is scheduled. Además, teachers may take off days for other reasons, with full or partial pay, without losing other employee beneﬁts. Although the rules covering such absences are both copious and 9. Over the period covered by our data, the number of school districts in the state was reduced through consolidation, de 119 a 117. Throughout, the consolidated district deﬁnitions are used to classify data by district. 118 l D o w n o a d e desde h t t p : / / directo . mi t . / / partícula alimentada – pdlf / / / / / 4 2 1 1 5 1 6 8 9 1 1 9 e d p 2 0 0 9 4 2 1 1 5 pd . . . . . f f b y g u e s t t o n 0 8 septiembre 2 0 2 3 Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor complicated, it is worth highlighting a few that apply to the most important types of absences—sick leave and vacation leave. Sick leave is credited to teachers in North Carolina at a ﬂat rate of one day per month of work. Signiﬁcantly for our study of teacher absence, this sick leave can accumulate in a teacher’s account indeﬁnitely and without limit, potentially giving some experienced teachers the ability to take very long spells of sick leave if necessary without losing either their teaching position or their employee beneﬁts. But another provision means that there are real costs to using this leave: at the time of a teacher’s retirement, any unused sick leave is converted to additional service credit and thus higher pension beneﬁts in the state’s deﬁned beneﬁt system. A teacher who has exhausted her store of sick leave can take as many as twenty additional days of extended sick leave but will be subject to a $50 reduction in salary per day.

A second category of absence, which we label personal leave, covers vol-
untary absences other than sick leave. A day of leave in this category entails a
deduction in pay, either $50 or a full day’s pay.10 A teacher who is absent due to illness but who has run out of accumulated sick leave and extended sick leave would be forced to use a type of leave in this category. The third category is vacation (or annual) leave. Unlike sick leave and per- sonal leave, which are by their nature usually unexpected and often disruptive to classroom instruction, vacation leave by design typically entails little dis- ruption of classroom routines. Most days taken in this category are in fact mandated to coincide with school vacation days, and teachers are not allowed to take other days off as vacation days without a principal’s permission. During their ﬁrst two years of service, teachers are credited with exactly the number of scheduled days of vacation during the school year that they earn (ten), thus leaving no additional days for them to use. In subsequent years, teachers earn more than this minimum, at rates that rise with experience.11 Combined with the ability to accumulate sick leave indeﬁnitely, this rising rate for vacation leave means that being absent from school is generally easier for teachers as they gain more years of experience. Como consecuencia, as we show below, aver- age absence rates due to sickness and vacation tend to rise with experience. The only signiﬁcant remaining category of leave is administratively mandated leave, usually for training and rarely held at a time that conﬂicts with classroom instruction. 10. One form of what we generically label here as personal leave has the ofﬁcial name of personal leave, which accumulates at a rate of 0.2 days per month of work. This form is subject to the $50 reducción
in salary per day.

11. The rate of accumulation for vacation leave rises from 1.15 a 2.15 per month, the top rate applying
to teachers with twenty or more years of experience. A maximum of thirty days of vacation leave
may be carried forward from one year to the next, with the excess converted to sick leave.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

119

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Mesa 1. Mean Absences by Type, 1994–95 to 2003–4

Año

Total

Sick Leave

Personal
Leave

Annual
Vacation Leave

Administrative

Type of Absence

1994–95

1995–96

1996–97

1997–98

1998–99

24.0

23.1

22.8

21.9

22.0

1999–2000

21.5

2000–1

2001–2

2002–3

2003–4

21.7

20.4

20.6

22.0

7.2

7.0

6.9

7.1

6.9

7.2

6.9

7.1

7.6

0.6

0.5

0.6

1.1

1.2

13.3

12.9

12.3

11.5

11.3

10.9

10.8

10.0

9.9

10.4

2.3

2.1

2.3

2.4

2.5

2.3

2.7

Notas: Table includes teachers working at least ten months. Sick leave includes sick leave,
extended sick leave, and sick leave bank; personal leave includes personal leave, ausencia
with deduction, absence without pay, voluntary shared leave, child involvement leave,
and other absence; annual vacation leave includes annual leave and annual leave for
catastrophic illness; administrative includes absences without deduction. See appendix
table A.1 for a complete list of leave categories.
Fuente: North Carolina Education Research Data Center.

Mesa 1 summarizes the main categories of leave taken by North Carolina
teachers between 1994–95 and 2003–4. For the period, sick leave averaged
7.1 days per teacher, for a rate of about 3.9 percent based on a 180-day school
año. Adding in personal leave, which averaged about 0.9 days over the period,
yields a slightly higher average rate of roughly 4.4 por ciento, a rate that is in the
same ballpark as the 5 percent suggested in the few previous studies of teacher
absences.12 Because they are usually not planned, absences in the sick leave
and personal leave categories are of paramount signiﬁcance for the functioning
of schools, and it is for this reason that we focus on them in the analysis in
succeeding sections of the article.

Under annual vacation leave, mesa 1 reveals a fairly steady decline, de
about thirteen days to about ten over the period. Based on our conversations
with school administrators, we believe this decline is largely an artifact of
differing and changing administrative practices. Since the ﬁrst ten days of
vacation leave correspond to mandatory school holidays, the amount charged
to a full-time teacher should never be less than ten, and indeed for most
districts the average number of vacation leave days is well above ten. But for
reasons not entirely clear to administrators in the affected districts or in the

12. See note 3.

120

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

state education department, the reported numbers for some districts are much
más bajo, in some cases covering only the excess vacation days above ten.13 For this
reason, our statistical analysis involving vacation leave in succeeding sections
examines variations only among teachers within the same district in the same
año, thus allowing for practices in this regard to differ across districts as well
as to change over time within each district. The last category of leave shown
in the table, administrative leave, shows little variation over time, although it
does differ across districts.14

As described above, the rules for accumulating leave explicitly favor more
experienced teachers. Not only can teachers carry forward some vacation leave
and all sick leave, the rate at which teachers earn vacation leave increases with
experiencia. The actual usage of leave reﬂects these factors, as illustrated by
the distribution of leave by experience level, shown in ﬁgure 1, for the 2000–1
school year. Novice teachers have the lowest average usage of sick leave, en
4.8 days a year, compared with over 8 days a year for teachers with ﬁve to
ten years of experience. Vacation leave also tends to rise with experience, en
least up to the last experience group. Excepting the most experienced teachers,
both personal leave and administrative leave are essentially constant across the
experience groups. To summarize, the most prominent regularity with respect
to experience is the markedly lower rate of sick leave for inexperienced teachers.
At the individual level, absence rates differ widely across the teacher work-
fuerza, and they also differ across schools and districts. To give an indication of
their variation at the individual level, figuras 2 y 3 present histograms for
days of absence per year for North Carolina public school teachers over the
years 1994–95 to 2003–4. Cifra 2, which shows the distribution of absences
for sick plus personal leave, reveals that the modal number of days of sick plus

13. Conversations with payroll and ﬁnance personnel in a number of districts and in the Department
of Public Instruction uncovered various explanations for large jumps in reported vacation leave,
including the adoption of new software, related changes in recordkeeping (including keeping parallel
records over a period leading to duplicate leave records), an agreement to compensate teachers with
annual vacation leave in return for working at school athletic events, and the practice in a few districts
of recording only days of annual leave above the ten mandated days. These practices all appeared to
be consistent with state policy, although they resulted in reported numbers of annual vacation leave
days apparently inconsistent with state policy. Sin embargo, the extent to which otherwise similar
districts differed in their average annual leave amounts could not be fully explained by any ofﬁcial.
14. For a detailed list of all leave categories and their frequency in one year, see appendix table A.1.
Although the data available to us are very rich, we had to make a few adjustments to deal with
several imperfections in the administrative records. Primero, records of absences for some teachers
in some years were missing. We drop these teacher-year observations. For the years 1994–95 to
2000–1, these observations accounted for fewer than 1.5 percent of the total, but the percentage of
missing teachers rises to 3.4 por ciento, 10.1 por ciento, y 10.9 percent in the years 2001–2 to 2003–4.
Segundo, we dropped the few observations with negative days of total leave or sick plus personal leave
(most likely reﬂecting adjustments to a previous year’s record) and those showing more than 150
días. Because they affected such small numbers of teachers, none of these adjustments made any
appreciable difference in results.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

121

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Cifra 1. Distribution of Absences by Years of Experience, 2000–1

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Cifra 2. Frequency Distribution, Sick and Personal Leave

personal leave over the period was zero. The median was six, and the mean was
eight. A very small number of teachers accounted for a disproportionate share
of the total days taken: el 10 percent of teachers showing the most days of sick
plus personal leave accounted for one-third of all teacher-year observations.

122

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Cifra 3. Frequency Distribution, Annual (Vacation) Leave

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

Cifra 4. 25th Percentile, Significar, and 75th Percentile for Sick Plus Personal Leave Absences, por
District Rank

The frequency distribution for vacation leave, shown in ﬁgure 3, reveals
two peaks, one at six days and the other at ten, a result of the apparent anomaly
in the reporting or recording of annual leave noted in the previous section.
The higher peak reﬂects the standard practice whereby all teachers are auto-
matically charged with the ten days per year corresponding to school holidays,
with any excess discretionary annual leave days added on top. The lower peak
reﬂects the sizable number of districts that, at least in some years, averaged
signiﬁcantly fewer than the standard ten days per teacher.

Absence rates varied markedly across schools and, en un grado menor,
across school districts. Cifra 4 illustrates the variation across districts by ﬁrst

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

123

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Cifra 5. Plot of District Average Rate of Sick Plus Personal Leave Absences in 2000–1 and 2003–4

showing the average sick plus personal absences ranked from lowest to highest
in the state. Whereas ﬁfteen districts averaged fewer than eight absences per
maestro, seven districts at the top had more than eleven per teacher. To show
the amount of variation within districts, the ﬁgure also plots for each the 75th
and 25th percentile number of absences. No es sorprendente, most of the variation
across districts occurs at the top of the distribution. It is worth noting that such
rankings by district tend to be fairly stable over time. As illustration, figura 5
shows the scatter plot of average absence rates by district in school years three
years apart. The high intertemporal correlation suggested by the ﬁgure is also
indicated by a correlation coefﬁcient of .653.15 Needless to say, districts with
consistently high rates of absence, as well as problematic schools within dis-
tricts, constitute a legitimate source of concern if teacher absence disrupts the
educational process. Variations such as these across schools and districts point
to the need for more research on institutional and organizational factors that
may be associated with persistent differences. It may be especially important
to better understand the role of principals in controlling teacher absenteeism.

3. WHICH TEACHERS ARE ABSENT MOST OFTEN?
To understand what kinds of teachers tend to be absent most often, we es-
timated ordinary least squares (OLS) regressions with absences of different
types as the dependent variable. We pooled records over the nine school years
from 1994–95 to 2003–4. Including all classroom teachers working at least

15. These high absence districts are small and are located in the coastal plain and the far western

mountains.

124

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

ten months yields a sample of more than 492,000 observaciones. Covariates
used as explanatory variables include the teacher’s gender, carrera, edad, and years
of experience, information on the teacher’s education and teacher credentials,
and information on the teacher’s school and district. Because of the strong
indication, noted above, that the conventions for recording vacation leave were
not constant across districts or over time, regressions that included such leave
were estimated with district-by-year ﬁxed effects. For the sake of comparison,
regressions for sick plus personal leave were estimated with and without these
efectos fijos, the resulting estimates being very close to each other.

Mesa 2 presents a subset of the estimated coefﬁcients for these regres-
siones. Estimates in column 2.1 are taken from the regression for sick plus per-
sonal leave that includes year indicators as well as variables describing district,
escuela, and individual characteristics.16 In contrast, the remaining equations
in the table, because they employ district ﬁxed effects, omit unchanging district
measures as well as year indicators. With respect to demographic characteris-
tics, columna 2.1 shows ﬁrst that black and other nonwhite teachers took slightly
less sick and personal leave than white teachers. The association with age and
gender is modeled with a series of dichotomous age indicators and interaction
terms with gender. The pattern of estimates for these indicators (not shown in
the table) indicate that female teachers, like female workers in the workforce at
grande, are absent more often than men. Por ejemplo, the estimates imply that
female teachers averaged 3.2 more days of sick plus personal days of leave than
male teachers at ages twenty-ﬁve and thirty-ﬁve and 1.3 more at age forty-ﬁve.17
The experience indicators reveal a pattern that reﬂects the limited amount
of sick and personal leave that new teachers can accumulate. In their second
año, teachers took an average of 1.4 more days than they did in their ﬁrst.
This gap increased to 2.1 days in their third and fourth years but varied there-
después. Inexperienced teachers therefore had considerably fewer absences due
to illness and other personal reasons than those with at least several years’
experiencia. As we note below, inexperienced teachers also took less vacation
leave. Teachers who graduated from a college in a bordering state had slightly
more absences than other teachers, which might reﬂect occasional trips home.
Fewer days were taken by teachers with high test scores, with master’s degrees,
who had National Board certiﬁcation, or who had graduated from a very com-
petitive college. Teachers in schools with higher percentages of free lunch
recipients tended to have higher absence rates, and the same is true for the

16. Descriptive statistics for the variables used are in appendix table A.2, and the full set of estimated

17.

coefﬁcients corresponding to equation 2.1 is given in appendix table A.3.
Ichino and Moretti (2006, mesa 1, pag. 37) report that overall, female workers in the United States
average about three more days of illness-related absences than males, with a smaller difference for
unmarried and childless workers. In separate regressions by gender, these differences are reﬂected
in part in larger experience effects for women than men.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

125

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Mesa 2. Selected Coefﬁcients from Regressions Explaining Teacher Absences: Pooled Data for 1994–95
to 2003–4

2.1

2.2

2.3

2.4

Dependent Variable

District by Year
Fixed effects?

Sick and
Personal
Leave
No

Sick and
Personal
Leave

Sí

Annual
(Vacation)
Leave

Sí

Sick, Personal,
and Vacation
Leave

Sí

Negro

Other nonwhite

Experience

1 año

2–3 years

4–5 years

6–10 years

11–30 years

Encima 30 años

Graduated NC college

Graduated college in
state bordering NC

Teacher test score

Teacher has master’s

degree

National Board certiﬁed

maestro

Graduated “very

competitive” college

Middle school

High school

School % free lunch

× elementary school

School % free lunch
× middle school

School % free lunch
× high school

District % free lunch

−.415
(.130)

−.108
(.300)

1.430
(.101)

2.103
(.101)

2.566
(.112)

2.467
(.126)

1.799
(.160)

3.694
(.197)

.072
(.044)

.190
(.067)

−.318
(.019)

−.219
(.049)

−1.006
(.094)

−.224
(.053)

−.391
(.086)

−.804
(.077)

.206
(.181)

.825
(.255)

1.222
(.301)

−.507
(.127)

−.493
(.297)

1.419
(.101)

2.094
(.101)

2.518
(.112)

2.395
(.124)

1.693
(.155)

3.574
(.192)

.075
(.044)

.215
(.067)

−.316
(.019)

−.227
(.048)

−1.024
(.094)

−.223
(.052)

−.398
(.085)

−.835
(.076)

.167
(.175)

.745
(.249)

1.248
(.292)

.216
(.287)

.288
(.053)

−.126
(.122)

.736
(.041)

1.178
(.041)

1.522
(.046)

2.036
(.051)

2.531
(.065)

1.780
(.080)

.684
(.018)

.168
(.027)

−.026
(.008)

−.281
(.020)

−.095
(.038)

−.019
(.022)

−.006
(.035)

−.099
(.031)

.127
(.074)

−.008
(.104)

−.149
(.122)

−.026
(.140)

−.247
(.324)

2.196
(.109)

3.309
(.109)

4.086
(.121)

4.480
(.136)

4.311
(.173)

5.546
(.212)

.723
(.048)

.349
(.073)

−.343
(.021)

−.521
(.053)

−1.007
(.102)

−.215
(.057)

−.407
(.093)

−.899
(.083)

.328
(.195)

.808
(.276)

1.121
(.325)

126

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Mesa 2. Continuado.

Dependent Variable

District by Year
Fixed effects?

School % nonwhite

District % nonwhite

Mean of dependent

variable

Número de

observaciones

2.1

Sick and
Personal
Leave

2.2

Sick and
Personal
Leave

Sí

2.3

2.4

Annual
(Vacation)
Leave

Sí

Sick, Personal,
and Vacation
Leave

Sí

.503
(.154)

−.438
(.221)

.036

8.68

.608
(.158)

.063
(.064)

.630
(.170)

.034

8.68

.064

11.31

.043

19.98

498,825

498,660

498,256

Notas: Standard errors in parentheses. Other variables in regression speciﬁcations included whether
teacher was male, teacher age indicators, interaction term between male and teacher age, log of
ratio of teacher salary to other teacher salaries within 30 miles of district, log of teacher salary, log of
ratio of teacher salary by teacher experience/salary type to nonteacher salaries for counties within
30 miles of district, whether graduated from college deemed competitive according to Barrons
ranking, county unemployment rate, and black teacher/nonwhite student percentage and other
nonwhite teacher/nonwhite student percentage interaction terms. Ecuación 2.1 also includes the
log of growth in district enrollment between 1990 y 1995 and the log of district enrollment
indicators for school year along with rural district and mountain and coastal region indicators. Para
the full list and means of equation 2.1 variables, see appendix table A.2. The full set of estimated
coefﬁcients corresponding to equation 2.1 is given in appendix table A.3. Equations 2.2–2.4 use
district-year ﬁxed effects.
Fuentes: North Carolina Education Research Data Center; authors’ calculations.

district free lunch percentage, but none of the estimated effects is very large
on its own. Por ejemplo, the coefﬁcient for proportion eligible for free lunch
in high schools implies that increasing a school’s free lunch share from the
25th to the 75th percentile would increase average absences for all the school’s
teachers by only about one-ﬁfth of a day per year. The estimated effects for
middle school and elementary schools are smaller.18 The district free lunch
percentage has almost no effect on absences. As for proportion nonwhite, el
school’s proportion is positively associated with absences while the district’s is
negatively correlated. Although these estimates seem to give little reason for
concern that low-income or minority students are experiencing very elevated
rates of teacher absence, it will be important to examine the actual incidence
of high absence rates, as we do in the next section.

The table’s remaining three regressions employ district-by-year ﬁxed ef-
efectos, which means that all coefﬁcients reﬂect only differences within districts
in a given year. Ecuación 2.2, which is directly comparable to equation 2.1,
produces very similar estimated coefﬁcients. This similarity suggests that

18. For the 25th and 75th percentiles at each level, ver tabla 3 notas.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

127

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

unaccounted-for differences across districts or over time are not important
in explaining variation in sick plus personal days. Ecuación 2.3, which ex-
amines annual vacation leave, reveals several contrasts with the comparable
equation for sick plus personal leave. One difference is a reduction in the
experience gradient. Teachers with two to three years of experience took 1.2
vacation days more than novice teachers, comparado con 2.1 more sick plus
personal days. Because vacation leave accumulates at a rising rate, more is
used in later years, until after thirty years, when it is presumably being saved
to augment retirement beneﬁts. Además, in contrast to equation 2.2, North
Carolina college graduates took an extra half day of vacation and National
Board certiﬁed teachers took an extra day. The effects of free lunch and racial
composition virtually disappear in explaining vacation leave.

In order to arrive at a comprehensive measure of absences most likely
associated with lost teaching time, we summed all absences not associated with
administrative reasons, shown in equation 2.4. By and large, the coefﬁcients
in this model reﬂect the patterns seen in equation 2.2.

4. DISTRIBUTIONAL ASPECTS OF TEACHER ABSENCE
Do teacher absences occur more frequently in schools serving low-income stu-
abolladuras? En ese caso, absences would join the list of unfavorable school characteristics
that disproportionately affect disadvantaged students, such as having inexperi-
enced teachers.19 Our data from North Carolina indicate that teacher absences
do indeed have this kind of distributional impact. As shown in equations 2.2
y 2.4 in table 2, otherwise similar teachers have slightly higher rates of
absence when they teach in schools and districts where higher percentages of
students are eligible to receive free lunches; equation 2.1 implies district-level
differences as well but with racial composition going in the opposite direction.
But it is not obvious that these regression results necessarily imply higher ab-
sence rates for low-income schools because the regressions also indicate lower
absence rates for inexperienced teachers, whom we know from previous work
to be more numerous in these same kinds of schools and districts (Clotfelter,
muchacho, y vigdor 2005; Clotfelter et al. 2007). To understand the full distri-
butional impact, por lo tanto, one needs to compare actual rates of incidence by
income level.

Mesa 3 presents tabulations showing for one year the average number
of absences taken by teachers in schools falling into each quartile of schools
deﬁned by free lunch percentage. To allow for the differences in take-up of
free lunch by school level, the averages are shown separately for each school

19. For empirical studies of distributional patterns of school resources, see Betts, Rueben, y

Danenberg (2000) and Clotfelter, muchacho, y vigdor (2005).

128

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Mesa 3. Average Absences per Teacher, Selected Categories, by School Income Quartile, 2000–1

Lowest Income
Quartile

2nd Quartile

3rd Quartile

Highest Income
Quartile

Elementary schools

FTE teachers

All absences

Sick leave

Personal leave

9,946

23.8

8.5

1.9

Annual vacation leave

10.7

Middle schools

FTE teachers

All absences

Sick leave

Personal leave

4,439

23.0

8.2

1.9

Annual vacation leave

10.3

High schools

FTE teachers

All absences

Sick leave

Personal leave

5,827

22.9

7.5

1.8

Annual vacation leave

10.6

10,187

23.2

8.0

1.9

10.7

4,507

22.1

7.8

1.6

10.4

5,743

20.4

6.6

1.5

9.8

9,930

23.8

8.2

1.8

11.3

4,512

22.3

7.6

1.5

10.6

5,744

21.6

6.6

1.5

11.2

9,597

23.4

7.8

1.8

11.2

4,449

22.3

7.3

1.5

10.8

5,844

21.3

6.5

1.5

10.8

Notas: FTE = equivalente a tiempo completo. Schools are classiﬁed by quartiles of percent free lunch, dónde
the quartile breaks are deﬁned separately for elementary, middle, and high schools, taking all years
together. Income quartiles are based on FTE teacher counts. Where data on percent free lunch
were missing, data for the school in the previous year or the following year were used instead,
where possible. All remaining schools were dropped. The lowest income quartile refers to schools
with the highest percentage of students eligible for free lunch. For the state’s schools, the 25th,
50th, and 75th percentiles for proportion free lunch were, respectivamente .244, .360, y .516 en
elementary schools; .204, .298, y .428 in middle schools; y .103, .169, y .268 in high
escuelas. Absences are for ten-month year or year equivalent.

nivel. For each level, the average number of teacher sick days is highest in
the bottom income quartile and lowest in the most afﬂuent quartile. El
differences between top and bottom quartiles in mean sick days is on the order
of one day per teacher or less. Asimismo, personal leave tends to be highest in
low-income schools, but the differences across the income spectrum are not
grande. A diferencia de, annual vacation leave tends to rise with income quartile
(recordar, sin embargo, that reporting conventions differ across districts). Porque
more afﬂuent schools tend to have more experienced teachers, and those
teachers have more access to annual leave days, this result is not surprising. En
any case, one would expect that days of annual leave do not carry with them the
same potential for lost instruction time that sick days do, since these absences

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

129

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Mesa 4. Prevalence of High-Absence Schools, by Income Quartile, Combined Years, 1994–95 to
2003–4

Lowest Income
Quartile

2nd Quartile

3rd Quartile

Highest Income
Quartile

Elementary schools

33.9

Middle schools

High schools

25.6

16.0

24.7

15.4

5.3

22.3

10.4

1.8

19.4

7.8

3.2

Notas: Schools are classiﬁed by quartiles of percent free lunch, where the quartile breaks are
deﬁned separately for elementary, middle, and high schools, taking all years together. (Ver
notas, mesa 3.) Table entries indicate weighted percentage of schools by income quartile that
were in the highest quartile of average absences, taking all years together (más que 9.84
sick + personal days per teacher), in at least ﬁve years in the period 1994–95 to 2003–4,
where the weights are FTE teachers times years in the sample. A school that appears in
different income quartiles over the period will be reﬂected according to the number of years
and FTE teachers in each income quartile. Schools appearing in fewer than ﬁve years were
omitted.

require a principal’s approval and are thus most often taken during teacher
workdays rather than on school days.

If teacher absences are harmful to learning, they are apt to be especially
damaging if they are schoolwide and occur year after year. En efecto, persistently
high absenteeism appears to be one hallmark of troubled schools.20 For this
reason, we sought to ﬁnd out if some schools in our sample tend to experience
consistently high rates of absence and, if so, whether those schools serve low-
income students. We looked at schools that had been, in at least ﬁve of the ten
years covered by our data, in the highest quartile of average sick plus personal
días. Out of 2,094 schools that were in our data for at least ﬁve years, 559
qualiﬁed by this criterion.

Mesa 4 shows the prevalence of high-absence schools by schools’ income
quartile over this period. Each cell in the table gives the percentage of schools,
by income level, that fell into the highest absence quartile.21 The table shows,
ﬁrst of all, that elementary schools are more likely to fall into the high-absence
category than are middle schools and that high schools are least prone among
all levels. This difference reﬂects in part the higher average rates of sick leave
in elementary schools, as shown also in table 3, but it could also be a reﬂection
of the smaller size of elementary schools, and thus their tendency for wider
swings in average absence rates from year to year. Within each level, sin embargo,
the pattern is quite clear, with high-absence schools being much more preva-
lent among those with low-income students than those serving student pop-
ulations with higher average family incomes. Por ejemplo, whereas a quarter

20. Ver, Por ejemplo, Dell’Angela and Little (2006) and Imants and Van Zoelen (1995).
21. Percentages are weighted by full-time equivalent teachers. See table 4 for a detailed description of

the calculations.

130

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

of middle schools fell into the persistently high absence category, fewer than
one in twelve middle schools serving the most afﬂuent quarter of students had
such consistently high rates of absence. En suma, low-income students in North
Carolina face an appreciably higher chance than afﬂuent ones of attending a
school with persistently high rates of teacher absence.

5. ABSENCES AND STUDENT ACHIEVEMENT
Common sense suggests that teacher absences will impede students’ academic
progress. To see if the data are consistent with this reasoning, we estimated
variants of a standard value-added model of the form:

Ait = a Ait-1 + b Absit + c Xit + uit,

(1)

where Ait is student i’s achievement test score in year t (normalized with mean
zero and unit standard deviation), Absit is the number of sick plus personal
days taken by student i’s teacher in year t, Xit is a vector of student, escuela,
and teacher characteristics, uit is an error term, y un, b, and c are estimated
coefﬁcients or vectors of coefﬁcients. In this model, the coefﬁcient on number
of sick plus personal days taken by a student’s teacher, b, measures the average
difference in achievement between otherwise similar students whose teachers
differed by one in the number of sick and personal days absent.

We used OLS to obtain the initial estimates reported in model A of table 5.
The richness of the available administrative data made it possible for us to
match most North Carolina students in grades 4 y 5 to the classroom teachers
who taught them math and English. This matching enabled us to compare the
academic achievement of students whose teachers differed in the number of
sick days taken, holding constant a long list of other student, maestro, y
school characteristics as well as the student’s previous achievement score.
As shown in the table, for math achievement, the coefﬁcient for the absence
variable is −0.0023 (s.e. = 0.0001). This ﬁnding implies that having a teacher
with ten additional sick days in a year would be associated with a reduced math
test score of about 2.3 percent of a standard deviation.22 By comparison, este
effect is slightly larger than that of changing schools and about half the size of
the effect of being eligible for the subsidized lunch program. For reading, el
coefﬁcient is less than half as large, implying that the same ten-day increase
in sick days would be associated with a lower test score of about 1 por ciento de un
standard deviation.

The coefﬁcients that emerge from this simple OLS model, sin embargo, son
likely to be biased. One possibility is that teacher absences may be corre-
lated with unmeasured aspects of teacher ability or effort, which would cause

22. The complete set of estimates is given in appendix table A.4.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

131

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Mesa 5. Teacher Absences and Student Achievement: Basic Estimates and Validity Checks (Es-
timated Coefﬁcients for Sick Plus Personal Days in Equations Explaining Normalized End-of-Grade
Tests, Grades 4–5, 1994–95 to 2003–4)

Matemáticas

Reading

Matemáticas

Reading

No ﬁxed effects

Teacher ﬁxed effects

−.0023∗∗∗
(.0001)

−.0011∗∗∗
(.0001)

−.0017∗∗∗
(.0001)

−.0009∗∗∗
(.0001)

−.0010∗∗∗
(.0002)

−.0030∗∗∗
(.0002)

Yes∗∗∗

−.0020∗∗∗
(.0002)

−.0025∗∗∗
(.0001)

Yes∗∗

−.0003∗
(.0002)

−.0015∗∗∗
(.0001)

Yes∗∗∗

−.0005∗∗∗
(.0002)

−.0013∗∗∗
(.0001)

Yes∗∗∗

−.0007∗∗∗
(.0002)

−.0023∗∗∗
(.0001)

−.0004∗∗
(.0002)

−.0012∗∗∗
(.0001)

Yes∗∗∗

−.0017∗∗∗
(.0002)

−.0018∗∗∗
(.0001)

−.0006∗∗∗
(.0002)

−.0010∗∗∗
(.0001)

Yes∗∗

Model A

Number of days absent

Model B

Number of days absent,

July–December

Number of days absent,

January–June

Coefﬁcients signiﬁcantly

diferente?

Model C

Number of absences covered

by certiﬁed substitute

Number of absences covered
by uncertiﬁed substitute

Coefﬁcients signiﬁcantly

diferente?

No. of observations

997,408

1,005,380

1,123,603

1,131,781

Nota: Standard errors in parentheses.
∗signiﬁcant at 10%; ∗∗signiﬁcant at 5%; ∗∗∗signiﬁcant at 1%
Fuente: North Carolina Education Research Data Center.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

omitted variables bias in the coefﬁcient of absences. En este caso, the absence
variable would reﬂect the combined effects of otherwise unmeasured charac-
teristics of teachers correlated with high absence rates and the effect of the
absences themselves. A second possibility that would threaten the validity of
OLS estimates is if absences are inﬂuenced by students’ performance, de este modo
subjecting OLS to simultaneity bias. Teachers whose students are struggling
academically, Por ejemplo, might tend to take more sick days out of frustration
or discouragement.

To account for the possibility of bias due to omitted variables correlated with
absences, we followed the approach of Miller, Murnane, and Willett (2007)
by estimating an alternate speciﬁcation using teacher ﬁxed effects. Tal
approach depends entirely on variation over time in a teacher’s absences to
estimate the relationship between absences and student achievement. adi-
tive time-invariant teacher characteristics, which could include unmeasured
ability or effort, are swept away and thus cannot lead to omitted variables bias.

132

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

These equations with teacher ﬁxed effects yielded somewhat smaller, aunque
statistically signiﬁcant, coeficientes. These ﬁxed effects models imply that ten
additional days of teacher absence would be associated with a decline of 1.7
por ciento de una desviación estándar (s.d.) in math achievement and 0.9 por ciento
standard deviation in reading.23 These magnitudes are less than the 3.3 por ciento
obtained by Miller, Murnane, and Willett (2007) for math but are of the same
order of magnitude. By way of comparison, our basic ﬁxed effect achievement
regression implies that having a teacher with 1–2 years experience is associ-
ated with higher achievement of 7.7 percent standard deviation in math and
4.6 percent in reading.24

The endogeneity problem is more difﬁcult to address. In the absence of
a good instrumental variable to deal with this problem, we adopted one addi-
tional strategy in an effort to pin down the causal effect of teacher absences
on student achievement. If teacher absences depress student learning, we rea-
soned, some absences might cause more damage than others. En particular,
teacher absences that occur early in the year would probably be less harmful
than those occurring in the second half, in the run-up to the annual testing
period near the end of school. Another possibility we considered is that ab-
sences that were covered by a certiﬁed substitute might be less harmful than
those covered by an uncertiﬁed substitute.

We therefore estimated two variants of the achievement model described
arriba. The ﬁrst variant divided teacher absences according to the month they
occurred—July to December and January to June. As shown in table 5, model B,
the estimated coefﬁcients of teacher absence differ signiﬁcantly between the
ﬁrst and second semesters, with the second semester effects being about three
times as large as the ﬁrst semester effect in math. Although the imposition
of teacher ﬁxed effects reduces most of the estimated coefﬁcients, those for
the two semesters remain statistically different from each other. Estos resultados
are strongly suggestive of a causal link between teacher absence and student
achievement in elementary grades.

Por el contrario, we found much smaller differences between the effects of
absences when absences are divided by the type of substitute, as shown in
model C. A pesar de, as expected, absences covered by an uncertiﬁed substitute
were associated with larger declines in achievement than those covered by

23. Ten days is approximately one standard deviation in the absence measure.
24. All the estimates presented in the text and tables are based on speciﬁcations in which number
of sick days plus personal days of absence enter linearly. We explored other functional forms,
including quadratic, square root, and discrete indicators for ranges of absences, but the implied
effects were very close to those due to the linear speciﬁcation.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

133

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

certiﬁed substitutes, the differences are statistically signiﬁcant in only one of
the two ﬁxed effects equations.25

If absences do indeed depress student achievement, it is natural to worry
about whether this effect might be more severe among certain, more vulnerable
estudiantes. En efecto, Molinero, Murnane, and Willett (2007) suggest that such a
difference in impact might explain the contrast in magnitudes between their
estimates and ours. We therefore estimated a series of equations of the basic
form of model A, agregando, seriatim, interaction terms that would indicate a
differential inﬂuence. These variants are shown in table 6. The ﬁxed effects
models show that the deleterious effect of teacher absences is greater among
students who are in rural districts; who are nonblack nonwhites, comparado
with whites (for math only); who are eligible for the free lunch program (matemáticas
solo, en el 10 solo nivel porcentual); and who scored below average the previous
año. Students whose parents did not graduate from high school or who were
taught by a teacher with two or fewer years of experience showed no statistically
signiﬁcant difference in effect of absences, except for the anomalous result that
those taught by inexperienced teachers actually gained a small amount in math
from additional teacher absences. Among these differential effects, the largest
was for low-achieving students. In the ﬁxed effect equation for math, ten days
of teacher absence would be associated with a drop of 3.3 por ciento de un estándar
deviation in score, compared with only 0.3 percent standard deviation for
above-average achievers.26

We believe that the teacher ﬁxed effect model goes a long way in dealing
with concerns that teacher absences are endogenous. To be sure, our method
is valid as long as there are no time-varying determinants of teacher absence
that correlate with unobserved determinants of student achievement. Semejante
a condition will be true in what we see as the likely case that year-to-year
variations in a teacher’s absences are driven by exogenous health effects rather
than the teacher’s response to that year’s class of students. Taken at face value,
the estimated coefﬁcients from our ﬁxed-effect model (model A in table 5)
imply that the achievement level for a student in grades 4–5 will fall .0017 de
a standard deviation in math and .0009 in reading for each day his or her
teacher is absent in the year. To put these effects in context, they imply, para
ejemplo, that ten additional days of absence would be associated with declines

25. As an additional check on the validity of our estimates, we sought to verify that our results were not
being driven by a comparatively small number of high-absence teachers. Thus we reestimated the
basic model A regressions, omitting teachers with more than ﬁfty absences in a year. La resultante
estimated coefﬁcients actually showed a somewhat larger effect for absences in math (–.0018 vs.
–0.0017 for the full sample, with ﬁxed effects) but no difference in reading.

26. It is worth noting that we found no statistically signiﬁcant difference in coefﬁcients by student

género.

134

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Mesa 6. Teacher Absences and Student Achievement: Heterogeneity (Estimated Coefﬁcients for Sick Plus
Personal Days and Interaction Terms in Equations Explaining Normalized End-of-Grade Tests, Grades 4–5)

A. Rural district

Absences

Interaction term

B. Black/other nonwhite student

Absence

Interaction—black

Interaction—other nonwhite

C. Students receiving free lunch

Absences

Interaction term

Matemáticas
(no ﬁxed
efectos)

Reading
(no ﬁxed
efectos)

Matemáticas
(ﬁxed
efectos)

Reading
(ﬁxed
efectos)

−.00202∗∗∗
(.00014)

−.00062∗∗∗
(.00019)

−.00084∗∗∗
(.00014)

−.00049∗∗∗
(.00014)

−.00124∗∗∗
(.00014)

−.00104∗∗∗
(.00021)

−.00064∗∗∗
(.00011)

−.00052∗∗∗
(.00017)

−.00217∗∗∗
(.00013)

−.00100∗∗∗
(.00010)

−.00171∗∗∗
(.00011)

−.00089∗∗∗
(.00009)

−.00032∗
(.00018)

−.00111∗∗
(.00044)

−.00009
(.00016)

.00028
(.00015)

−.00105∗∗∗
(.00040)

−.00065∗∗
(.00032)

.00006
(.00014)

−.00036
(.00033)

−.00213∗∗∗
(.00013)

−.00038∗∗
(.00015)

−.00099∗∗∗
(.00010)

−.00163∗∗∗
(.00012)

−.00086∗∗∗
(.00010)

−.00018
(.00014)

−.00022∗
(.00012)

−.00006
(.00013)

D. Parent without high school degree

Absences

Interaction term

−.00229∗∗∗
(.00012)

−.00105∗∗∗
(.00009)

−.00172∗∗∗
(.00011)

−.00087∗∗∗
(.00009)

−.00022
(.00027)

−.00036
(.00028)

−.00023
(.00024)

−.00026
(.00026)

mi. Students with below-average test score, lagged

Absences

Interaction term

−.00084∗∗∗
(.00013)

−.00300∗∗∗
(.00013)

.00057∗∗∗
(.00009)

−.00351∗∗∗
(.00013)

−.00028∗∗
(.00012)

−.00301∗∗∗
(.00012)

.00073∗∗∗
(.00010)

−.00352∗∗∗
(.00012)

F. Teachers with 0–2 years experience

Absences

Interaction term

−.00233∗∗∗
(.00013)

−.00112∗∗∗
(.00009)

−.00178∗∗∗
(.00011)

−.00092∗∗∗
(.00009)

.00014
(.00035)

.00039
(.00026)

.00051∗
(.00030)

−.00033
(.00026)

Number of observations

997,408

1,005,380

1,123,603

1,131,781

Nota: Standard errors in parentheses.
∗signiﬁcant at 10%; ∗∗signiﬁcant at 5%; ∗∗∗signiﬁcant at 1%

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

135

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

in achievement equal to about one-ﬁfth the advantage of having a teacher with
1–2 years experience, compared with having a novice teacher.27

6. CAN ABSENCES BE REDUCED THROUGH INCENTIVES?
Teacher absences are socially costly. Suppose we accept as valid the estimates
from the teacher ﬁxed effect models above. There are two ways to translate
these test scores into social costs. One would be to rely on estimates of the
relationship between test scores and lifetime earnings, or some other long-
range outcome. A simpler method would be to consider estimates of the cost
of offsetting these test score declines, based on existing interventions.28 Two
recent analyses, of the Tennessee STAR experiment and of a teacher retention
bonus program in North Carolina, suggest that the cost of increasing one
student’s test score in one subject by 1 percent of a standard deviation is in the
range of $36 a $39 in current dollars.29 Assuming a class size of twenty-ﬁve
students and that each teacher teaches both math and reading, the achievement
costs of a single absence are on the order of $250.30 Beyond these very rough calculations of the instructional costs of a teacher absence, school districts also face the cost of paying a daily wage rate to a substitute teacher, which could amount to as much as $90.31 The existence of both these types of costs
suggests that a policy of unlimited free absences for teachers would be socially
inefﬁcient.

From a policy perspective, it would be useful to know how teachers respond
to policies that impose some cost on the decision to take an absence. El
structure of teacher absence policy in North Carolina provides us with an
opportunity to address this question. Como se revisó anteriormente, teachers can take up
to ten sick days per year without penalty, with unused sick days being carried
over into subsequent school years. Whenever the supply of “free” sick days is
exhausted, teachers may take up to twenty additional sick days, at a cost of $50 27. For math, ten days of absence implies a reduction of .017 s.d., cual es 23 por ciento de la .0736 s.d. difference associated with having a teacher with 1–2 years experience. For reading, the comparable calculation is .009/.0467, o 19 por ciento. 28. This approach is reasonable under the presumption that interventions can be scaled upward or downward at constant average cost to deliver a precise dose of test score improvement. Given the potential for nonlinear dose response in most interventions, this assumption is clearly questionable. Our goal here is to provide a ballpark estimate of the instructional costs associated with the typical teacher absence, not to propose that any particular intervention be applied to students of an absent teacher. 29. See Krueger (1999) and Clotfelter et al. (2008), with details of these calculations in the latter. Using less conservative assumptions, the former study suggests an even larger cost estimate, on the order of $100.
(math coefﬁcient + reading coefﬁcient)∗25 students/class∗$36. In the 2006–7 school year, the Wake County Public School System, the state’s second largest at the time, paid a daily wage of $84 to certiﬁed substitute teachers. Adding the 7.65 percent employer’s
share of payroll taxes for Social Security and Medicare brings the cost to $90. 30. 31. 136 l D o w n o a d e desde h t t p : / / directo . mi t . F / / e d u e d p a r t i c e – pdlf / / / / / 4 2 1 1 5 1 6 8 9 1 1 9 e d p 2 0 0 9 4 2 1 1 5 pd . . . . F . f por invitado 0 8 septiembre 2 0 2 3 Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor per day. Appendix table A.1 indicates that about 15,000 of these extended sick days were taken in the 2000–1 school year. In that same year, teachers took over 535,000 free sick days. In any school year, a teacher’s supply of free sick days will depend on her experience level and on the number of sick days taken in prior years. Thus the impact of the $50 charge for extended sick days can be identiﬁed by comparing
teachers who have taken a comparable number of sick days in a given year,
exploiting the fact that some teachers will exhaust their supply of free days
earlier than others.

To analyze the impact of the $50 charge on teacher absences, we estimated a Cox proportional hazard model, analyzing a teacher’s decision to take sick day t conditional on having already taken t − 1 sick days in a given school year. Our estimated model takes the form: logit[λ(tijs)] = α + β1 Xij + β2 Xjs + β3 Ctij (2) where i indexes teachers, j indexes school years, and s indexes schools. The term λ(tijs) represents the conditional probability that t is the last sick day taken by a teacher in a given year, conditional on the fact that sick days 1 through t − 1 were not the last. The speciﬁcation controls for a vector of teacher characteristics, Xij, and school characteristics, Xjs. The independent variable of interest is the cost of sick day t for teacher i in year j, Ctij.32 The impact of the cost parameter on the dependent variable is identiﬁed by comparing teacher/year observations with identical values of t but different values of Ctij. Por ejemplo, an inexperienced teacher might start paying the penalty on the twelfth sick day, while more experienced teachers with a greater supply of banked sick leave would not face the penalty until many more days had been taken. On day twelve, por lo tanto, the cost of another sick day is $50 para el
inexperienced teacher but zero for the experienced one.

Mesa 7 presents the results of estimating equation 2 with data on 414,959
teacher/year observations between 1995 y 2005. The number of absences

32.

Idealmente, we would prefer to estimate a model that used the cost of sick day t + 1 as the independent
variable of interest. Desafortunadamente, we lack reliable data on the size of each teacher’s bank of available
free sick days. When we attempted to impute this information using the subsample teachers with
complete histories of employment and sick days taken, we were unable to accurately forecast which
teachers would be required to take extended sick days in a given year. There are a number of
conceivable reasons for these forecast errors. Teachers actually accrue sick days at the rate of one
per month rather than ten per year; sin embargo, our database is not sufﬁciently detailed to allow us
to observe the month in which a sick day was taken in all cases. Teachers also occasionally have
options to “borrow” sick days from other teachers, a practice that may not be fully documented in
our data. Our use of the cost of sick day t in place of the cost of sick day t + 1 implies that we have
some degree of errors-in-variables bias, which should lead us to understate the impact of monetary
incentives on absence taking. When we attempted to use our estimates of the cost of sick day t + 1
en cambio, we obtained coefﬁcients even more attenuated than the ones presented here.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

137

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Mesa 7. Absences and the $50 Absence Penalty: Cox Proportional Hazard Model Estimates Variable Cost Teacher characteristics (white omitted) Black Other nonwhite Experience (0 years omitted) 1 year 2–3 years 4–5 years 6–10 years 11–30 years Over 30 years Log of teacher salary Log of salary/alternate teacher salary ratio Log of salary/nonteaching teacher salary ratio Graduated NC college Graduated college in state bordering NC Teacher test score Teacher has master’s degree National Board certiﬁed teacher Graduated “very competitive” college Graduated “competitive” college School characteristics (elementary school omitted) Middle school High school School % free lunch × elementary school School % free lunch × middle school School % free lunch × high school School % student nonwhite District characteristics District % free lunch District % nonwhite County unemployment rate Black teacher × student nonwhite percentage Hazard Ratio (standard error) 1.003 (.0002) .972 (.011) 1.003 (.029) .775 (.009) .701 (.008) .672 (.008) .660 (.009) .672 (.011) .641 (.013) .941 (.033) 1.002 (.010) .847 (.011) 1.003 (.004) .987 (.006) 1.040 (.002) 1.059 (.005) 1.063 (.009) 1.052 (.005) 1.027 (.004) .985 (.008) 1.048 (.008) 1.000 (.0002) .999 (.0002) .999 (.0003) .914 (.014) .973 (.027) 1.027 (.022) .992 (.001) .960 (.018) Other nonwhite teacher × student nonwhite percentage .998 (.044) Average growth in district enrollment, 1990–94 Log of district enrollment Rural district .521 (.078) .998 (.002) .946 (.004) 138 l D o w n o a d e desde h t t p : / / directo . mi t . / / partícula alimentada – pdlf / / / / / 4 2 1 1 5 1 6 8 9 1 1 9 e d p 2 0 0 9 4 2 1 1 5 pd . . . F . . f por invitado 0 8 septiembre 2 0 2 3 Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor Table 7. Continuado. Variable Coastal district Mountain district School Year (1995 [1994–95] omitted) 1996 1997 1998 1999 2000 2001 2002 2003 2004 Log likelihood LR chi2 (147) Number of subjects Hazard Ratio (standard error) .992 (.005) 1.011 (.006) 1.030 (.008) 1.023 (.008) 1.045 (.008) 1.014 (.008) 1.043 (.008) 1.023 (.008) 1.048 (.009) 1.012 (.009) .980 (.008) −4,960,890.5 16,718.06 414,959 Nota: Variables for male, edad, and age interacting with gender were omitted. taken by these teachers over the time period was more than 2.6 millón. The table entries are hazard ratios, which carry a different interpretation than typ- ical regression coefﬁcients. An independent variable associated with a hazard ratio greater than one is a factor that makes it more likely that a teacher will stop taking absences after absence t, while variables associated with haz- ard ratios less than one make it less likely that a teacher will stop taking absences. The variable of interest, the cost in dollars of taking absence t, has a hazard ratio of 1.003, which is signiﬁcantly greater than one. To determine the impact of a $50 costo, this ratio needs to be raised to the 50th power. These results thus
indicate that associating a $50 penalty with sick day t increases the likelihood that no further sick days will be taken by the affected teacher in the given year by 16 por ciento, compared with a situation in which there was no cost to the teacher of taking an additional sick day. To evaluate the magnitude of this impact, suppose that the same effect would result from applying the $50 penalty to all sick days including the ﬁrst.
Suppose further that teachers take an average of seven sick days per year and
that the probability of taking sick day t, conditional on taking t − 1, is a constant
valor. Simple arithmetic shows that this constant value is approximately equal

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

139

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

a 0.875.33 The likelihood of sick day t being the last, conditional on taking sick
day t, is thus 0.125. Increasing this value by 16 por ciento, a 0.145, would reduce
the mean number of absences taken to 5.9.

The average teacher would be charged about $300 for absences in each school year. This sum could be offset by increasing base salaries. Districts could increase salaries still further by applying cost savings associated with the 1.1 averted absences per year. A revenue-neutral policy change, incorporating $100
in savings associated with averted payments to substitute teachers, would thus
increase teacher salaries by roughly $400 per year, in exchange for teachers accepting a $50 charge for each sick day taken.34 Districts willing to compensate
teachers for averted educational costs could push the base salary increase still
higher.35

The remaining teacher and school-level covariates in the hazard model each
display relationships with the decision to stop taking absences that are fully
consistent with the coefﬁcients in table 2, a correspondence that increases
our conﬁdence that the hazard model has identiﬁed a true deterrent effect
associated with the $50 penalty. 7. CONCLUSION Teacher absences are important for four main reasons. Primero, hiring substitute teachers not only costs money but also consumes valuable administrative re- sources, often in the mundane form of early morning phone calls by principals or assistant principals. In North Carolina, sick and personal leave represents slightly more than 4 percent of the standard 180-day school year, a rate typical in American public schools, though quite small in comparison with rates ob- served in other contexts. Although absence rates in teaching tend to be higher than ostensibly comparable ﬁgures for other similar occupations and sectors, sin embargo, they are not wildly out of line. De hecho, one could argue that it is precisely the opportunity to take the occasional day off that makes a teaching career attractive to many people with children. Except for schools and districts 33. Under the stated assumptions, referring to the conditional probability as p, the expected number of sick days taken is p + p2 + . . . + pT , where T is the length of the school year. If we replace this expression with an inﬁnite series (cid:5) pi , the expected value can be expressed as p/(1 – p). The term p/(1 – p) is equal to 7 when p = 7/8. The impact of extending the expected value to an inﬁnite series is negligible; if p = 7/8 then only one in 27.5 billion teachers will be expected to take as many as 180 sick days. 34. Politically, this policy would be easier to implement if advertised as a $400 bonus for perfect
attendance, reduced by $50 each day, with penalties accruing to teachers who took more than eight sick days in a given year. Nota, sin embargo, that risk-averse teachers might reject a policy that introduced the possibility of salary reductions even if their expected compensation increased. 35. Recall also that our estimate of the impact of the $50 penalty likely suffers from attenuation bias.
Thus the net savings in terms of absences averted would likely be higher than this estimate.

140

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

with persistently high rates of absence, entonces, the rate of teacher absences itself
should probably not be a cause for great concern.

The second reason to pay attention to teacher absences is their effect on
Logro estudiantil: when regular teachers are not in the classroom, oppor-
tunities for students to learn are cut short. This commonsense conclusion is
bolstered by statistical evidence showing that students whose teachers miss
more days for sickness score lower on state achievement tests. The third
reason to worry about teacher absences is that they occur with greater fre-
quency in low-income schools. Teacher absences therefore join other charac-
teristics of teachers that are distributed unequally across schools and should
be included in discussions of equity in the provision of public schooling.
The fourth reason to pay attention to teacher absences is that because the
demand for absences is price elastic, they can be inﬂuenced by school dis-
trict compensation policies. Our results suggest that teachers’ valuation of
a marginal sick day is in some cases less than $50, which is in turn less
than even the most conservative estimates of the marginal social cost of an
ausencia.

En general, entonces, policies that create or increase incentives to reduce the num-
ber of absences teachers take can be advocated on two fronts. From an efﬁciency
standpoint, these policies have the potential to simultaneously raise teachers’
expected compensation and reduce districts’ expected costs.36 From an equity
perspectiva, policies that reduce absences have the potential to reduce one of
the many resource disparities between high- and low-poverty schools.

Previous research suggests that policies regarding the number of absences,
the ability to carry forward unused sick days, the beneﬁts (if any) of not using
all allowable days, and school-level requirements about reporting absences
all have the potential to inﬂuence the actual rate of teacher absenteeism. En
assessing the desirability of adjusting such policies, policy makers must weigh
the costs of absences—budgetary, administrative, and educational—against
the degree to which more lenient policies might make teaching an attractive
career option.

We are grateful to Robert Malme, l. Patten Priestley, and Marco Hernandez for re-
search assistance, to Ronald Ehrenberg, Dave Marcotte, and Richard Murnane for
helpful comments, to the North Carolina Education Research Data Center for assis-
tance obtaining and using data for North Carolina public schools, and to the Spencer
Foundation and the National Center for Analysis of Longitudinal Data in Education
Investigación (CALDER) for ﬁnancial support. The views expressed are those of the authors
and do not necessarily reﬂect those of any institution.

36. Depending on the degree of risk aversion among teachers, the result may indeed represent a Pareto

mejora.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

141

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

REFERENCIAS
Banerjee, Abhijit, and Esther Duﬂo. 2006. Addressing absence. Journal of Economic
Perspectives 20: 117–32.

apuestas, Julian R., Kim S. Rueben, and Anne Danenberg. 2000. Equal resources, igual
resultados? The distribution of school resources and student achievement in California. san
Francisco: Public Policy Institute of California.

Bradley, Steve, Colin Green, and Gareth Leeves. 2005. Worker absence and shirking:
Evidence from matched teacher-school data. Labour Economics 14 (3): 319–34.

Chaudhury, Nazmul, Jeffrey Hammer, Michael Kremer, Karthik Muralidharan, y
F. Halsey Rogers. 2006. Missing in action: Teacher and health worker absence in
developing countries. Journal of Economic Perspectives 20: 91–116.

Clotfelter, Charles T., Elizabeth Glennie, Helen F. muchacho, and Jacob L. Vigdor. 2008.
Would higher salaries keep teachers in high-poverty schools? Evidence from a policy
intervention in North Carolina. Journal of Public Economics 92: 1352–70.

Clotfelter, Charles T., Helen F. muchacho, and Jacob L. Vigdor. 2005. Who teaches whom?
Race and the distribution of novice teachers. Revisión de la economía de la educación 24: 377–92.

Clotfelter, Charles T., Helen F. muchacho, Jacob L. Vigdor, and Justin Wheeler. 2007. High
poverty schools and the distribution of teachers and principals. North Carolina Law
Revisar 85: 1345–79.

Dell’Angela, Tracy, and Darnell Little. 2006. Teachers miss days; poor kids miss out:
Educators at some struggling schools take most time off, analysis shows. chicago
Tribune, 25 Septiembre.

Duﬂo, Esther, and Rema Hanna. 2005. Monitoring works: Getting teachers to come to
escuela. NBER Working Paper No. 11880.

Ehrenberg, Ronald G., Randy A. Ehrenberg, Daniel I. rees, and Eric L. Ehrenberg. 1991.
School district leave policies, teacher absenteeism, y rendimiento estudiantil. Diario
of Human Resources 26 (1): 72–105.

Glewwe, Pablo, Nauman Ilias, and Michael Kremer. 2003. Teacher incentives. NBER
Working Paper No. 9671.

graham, Kristen A. 2006. Teachers are truant, también, reform commission says. Philadel-
phia Inquirer, 7 December.

Hobbs, Tawnell D. 2002. DISD hopes to cut daily average of 678 teachers missing
escuela. Dallas Morning News, 24 Septiembre.

Ichino, Andrea, and Enrico Moretti. 2006. Biological gender differences, absenteeism
and the earning gap. NBER Working Paper No. 12369.

Imants, Jeroen, and Ad Van Zoelen. 1995. Teachers’ sickness absence in primary
escuelas, school climate and teachers’ sense of efﬁcacy. School Organization 15: 77–86.

Kossan, Pat. 2006. School districts, students paying price for teacher absences. Arizona
República, 27 Enero.

142

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

krüger, Alan B. 1999. Experimental estimates of education production functions.
Revista trimestral de economía 114: 497–532.

Molinero, Raegen T., Richard J. Murnane, and John B. Willett. 2007. Do teacher absences
impact student achievement? Longitudinal evidence from one urban school district.
NBER Working Paper No. 13356.

Podgursky, Miguel. 2003. Fringe beneﬁts. Education Next 3: 71–76.

Roza, Marguerite. 2007. Frozen assets: Rethinking teacher contracts could free bil-
lions for school reform. Education Sector Report. Available www.educationsector.org/
usr doc/frozenassets.pdf. Accedido 2 Julio 2008.

A NOSOTROS. Bureau of Labor Statistics. 2006. Household data annual averages: Mesa 47. Avail-
able www.bls.gov/cps/cpsaat47.pdf. Accedido 2 Julio 2008.

Cuadro A.1. Absence Codes, 2000–1 School Year

Code/Deﬁnition

01 Sick leave, NS

02 Voluntary shared leave

03 Extended sick leave, NS

Número

474,659

1,415

13,494

04 Absence without deduction, NS

186,445

05 Absence with deduction, NS

06 Personal leave, NS

07 Absence without pay, NS

08 Sick leave bank, NS

10 Child involvement leave

11 Sick leave, CS

12 Other absence

13 Extended sick leave, CS

14 Absence without deduction, CS

15 Absence with deduction, CS

16 Personal leave, CS

17 Absence without pay, CS

18 Sick leave bank, CS

20 Annual leave

22 Annual leave for catastrophic illness

28 Bonus annual leave

No code deﬁned

705

53,707

46,868

642

62,683

1,633

28,064

172

9,061

2,311

644,166

Percentage

Classiﬁcation

31.1

0.1

0.9

12.2

0.1

3.5

3.1

0.0

4.1

0.0

0.1

1.4

0.0

0.6

0.2

0.0

42.2

0.0

PAG

Notas: Observations are at the teacher, absence code, and pay period levels. Observations can
include duplicate entries for teachers teaching in more than one school.
NS = uncertiﬁed substitute; CS = certiﬁed substitute.
Absence classiﬁcation: S = sick leave; P = personal leave; V = vacation (annual) leave; A =
administrative leave.

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

143

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Tabla A.2. Means of Teacher Absence Regression Variables, Pooled Data for 1994–95 to
2003–4

Variable

Male

Negro

Other nonwhite

Teacher age

Teacher age × male

Experience (zero years omitted)

1 año

2–3 years

4–5 years

6–10 years

10–30 years

Encima 30 años

Log of salary

Estándar

Mean Deviation Min Max

0.178

0.149

0.017

0.383

0.356

0.128

39.035

10.868

6.615

15.076

0.053

0.089

0.077

0.152

0.495

0.059

0.223

0.285

0.266

0.359

0.500

0.236

10.514

0.176

10.156 10.899

Log of salary/alternate teacher salary ratio

Log of salary/nonteaching salary ratio

Graduated NC college

Graduated college in state bordering NC

Teacher test score

Teacher has master’s degree

National Board certiﬁed teacher

Graduated “very competitive” college

Graduated “competitive” college

Middle school

High school

0.047

0.150

0.729

0.084

0.038

0.295

0.025

0.179

0.550

0.217

0.269

0.456

0.155

0.384

0.497

0.413

0.443

School % free lunch × elementary school

School % free lunch × middle school

School % free lunch × high school

19.684

24.254

7.037

15.663

5.389

11.582

School % student nonwhite

District % free lunch

District % student nonwhite

County unemployment rate

Black teacher × student nonwhite percentage

39.3

30.6

38.4

4.987

0.086

Other nonwhite teacher × student nonwhite percentage

0.010

25.3

12.1

19.2

2.036

0.228

0.083

0.233

−0.562

0.590

0.236

−0.514

0.783

0.445

0.277

0.954 −28.194

3.743

99.140

94.866

97.902

100

75.868

0.823 97.418

1.2

18.2

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Growth in district enrollment from previous year

Log of district enrollment

Rural district

Coastal district

0.014

9.827

0.435

0.170

0.014

−0,028

0.999

1.087

0.496

0.376

6.498 11.724

144

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Tabla A.2. Continuado.

Variable

Mountain district

School year (1995 [1994–95] omitted)

1996

1997

1998

1999

2000

2001

2002

2003

2004

Means of dependent variables:

Sick + personal days

Leave (vacation) días

Sick + personal + vacation days

Number of observations

Significar

Estándar
Deviation

0.214

0.410

0.284

0.289

0.294

0.299

0.303

0.306

0.308

0.311

0.321

10.959

5.865

12.945

0.089

0.092

0.096

0.099

0.102

0.105

0.106

0.108

0.116

8.675

11.309

19.982

499,462

mín.

máx.

150

Fuentes: North Carolina Education Research Data Center; authors’ calculations.

Table A.3. Full Regression Explaining Teacher Absence Due to Sickness and Personal Leave: Pooled
Data for 1994–95 to 2003–4

Variable

Características del docente

Male

Negro

Other nonwhite

Experience (zero years omitted)

1 año

2–3 years

4–5 years

6–10 years

10–30 years

Encima 30 años

Teacher salary

Salary/alternate teacher salary ratio

Salary/nonteaching salary ratio

Graduated NC college

Graduated college in state bordering NC

Coefﬁcient (standard error)

1.230 (18400.9)

−.507 (.127)

−.493 (.297)

1.419 (.101)

2.094 (.101)

2.518 (.112)

2.395 (.124)

1.693 (.155)

3.574 (.192)

.476 (.362)

−.145 (.105)

1.384 (.136)

.075 (.044)

.215 (.067)

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

145

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Table A.3. Continuado.

Variable

Teacher test score

Teacher has master’s degree

National Board certiﬁed teacher

Graduated “very competitive” college

Graduated “competitive” college

School characteristics

Elementary school (omitted)

Middle school

High school

School % free lunch × elementary school

School % free lunch × middle school

School % free lunch × high school

School % student nonwhite

Características del distrito

District % free lunch

District % student nonwhite

County unemployment rate

Black teacher × student nonwhite percentage

Other nonwhite teacher × student nonwhite percentage

Coefﬁcient (standard error)

−.316 (.019)

−.227 (.048)

−1.024 (.094)

−.223 (.052)

−.061 (.041)

−.398 (.085)

−.835 (.076)

.167 (.175)

.745 (.249)

1.248 (.292)

.503 (.154)

.216 (.287)

−.438 (.221)

−.028 (.013)

.304 (.207)

.465 (.460)

Growth in district enrollment from previous year

−10.060 (1.707)

Log of district enrollment

Rural district

Coastal district

Mountain district

School year (1995 [1994–95] omitted)

1996

1997

1998

1999

2000

2001

2002

2003

2004

Mean of dependent variable

Number of observations

.102 (.024)

.280 (.047)

−.035 (.057)

−.366 (.061)

−.281 (.079)

−.202 (.077)

−.406 (.077)

−.166 (.078)

.191 (.080)

.516 (.087)

.130 (.090)

.379 (.090)

.781 (.084)

.036

8.68

498,825

Notas: Standard errors in parentheses. Age and age-gender indicators omitted.
Fuentes: North Carolina Education Research Data Center; authors’ calculations.

146

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Table A.4. Regression Estimates Explaining Normalized Achievement Test Scores, Grades 4 y 5,
1995–2004

Ecuación

Male

Negro

Hispano

Otra raza

Age in grade 3

Parents are college graduates (omitted)

Parents are high school graduates

Parents are high school dropouts

Limited English

Gifted

Special needs

Subsidized lunch

Repeating grade

Lagged math or reading score

School change

Structural school change

Classroom characteristics

Class size

Percent nonwhite

Percent subsidized lunch

Percent college grad (omitted)

Percent high school graduates

Percent high school dropouts

OLS
MATH

3A.1

OLS
READING MATH

Maestros
Fixed Effects Fixed Effects

Maestros

READING

3A.2

3A.3

3A.4

0.0098
(0.0020)∗∗

−0.0276
(−0.0021)∗∗

0.0110
(0.0017)∗∗

−0.0815
(0.0022)∗∗

−0.1208
(0.0021)∗∗

−0.0951
(0.0018)∗∗

0.0765
(0.0042)∗∗

0.0522
(0.0041)∗∗

0.0384
(0.0040)∗∗

−0.0712
(0.0037)∗∗

0.0578
(0.0036)∗∗

0.0389
(0.0031)∗∗

−0.0273
(0.0019)∗∗

−0.1272
(0.0020)∗∗

0.0385
(0.0039)∗∗

−0.0080
(0.0033)∗

−0.0588
(0.0014)∗∗

−0.0427
(0.0014)∗∗

−0.0586
(0.0012)∗∗

−0.0440
(0.0013)∗∗

−0.1051
(0.0013)∗∗

−0.1077
(0.0014)∗∗

−0.1041
(0.0013)∗∗

−0.2100
(0.0028)∗∗

−0.2447
(0.0030)∗∗

−0.2111
(0.0026)∗∗

0.0120
(0.0058)∗

−0.0434
(0.0061)∗∗

0.0069
(0.0050)

0.2698
(0.0020)∗∗

0.2112
(0.0017)∗∗

0.2748
(0.0017)∗∗

−0.1325
(0.0020)∗∗

−0.1692
(0.0022)∗∗

−0.1367
(0.0018)∗∗

−0.0475
(0.0013)∗∗

−0.0584
(0.0014)∗∗

−0.0488
(0.0012)∗∗

0.5403
(0.0047)∗∗

0.7237
(0.0009)∗∗

0.4674
(0.0048)∗∗

0.7031
(0.0009)∗∗

0.4503
(0.0047)∗∗

0.7214
(0.0008)∗∗

−0.1064
(0.0014)∗∗

−0.2409
(0.0029)∗∗

−0.0524
(0.0057)∗∗

0.2185
(0.0016)∗∗

−0.1753
(0.0021)∗∗

−0.0599
(0.0013)∗∗

0.3964
(0.0053)∗∗

0.7001
(0.0008)∗∗

−0.0228
(0.0018)∗∗

−0.0153
(0.0018)∗∗

−0.0161
(0.0016)∗∗

−0.0108
(0.0017)∗∗

−0.0464
(0.0053)∗∗

−0.0438
(0.0042)∗∗

−0.0092
(0.0046)∗

−0.0077
(0.0041)

−0.0036
(0.0003)∗∗

−0.0024
(0.0002)∗∗

−0.0050
(0.0003)∗∗

−0.0031
(0.0002)∗∗

−0.0051
(0.0061)

0.0071
(0.0045)

−0.0131
(0.0086)

−0.0383
(0.0083)∗∗

−0.0027
(0.0060)∗∗

−0.0516
(0.0076)∗∗

−0.0157
(0.0074)∗

−0.0129
(0.0065)∗

−0.0443
(0.0068)∗∗

−0.0303
(0.0050)∗∗

−0.0272
(0.0075)∗∗

−0.0182
(0.0064)∗∗

0.0027
(0.0148)

−0.0289
(0.0110)∗∗

−0.0322
(0.0146)∗

−0.0017
(0.0127)

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

147

ARE TEACHER ABSENCES WORTH WORRYING ABOUT?

Table A.4. Continuado.

Ecuación

Lagged class average math score

Características del docente

Male

Negro

Hispano

Otra raza

Same race as the student

Same gender as the student

Teacher credentials

No experience (omitted)

1–2 years

3–5 years

6–12 years

13–20 years

21–27 years

28+ años

Regular license (omitted)

Lateral entry

Interact continuing/lateral entry

Other license

Master’s degree

National Board certiﬁed

Undergraduate college noncompetitive

(omitted)

Competitive

148

OLS
MATH

3A.1

OLS
READING MATH

Maestros
Fixed Effects Fixed Effects

Maestros

READING

3A.2

3A.3

3A.4

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d

b
y
gramo
tu
mi
s
t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

0.0265
(0.0035)∗∗

0.0593
(0.0027)∗∗

−0.0322
(0.0034)

0.0178
(0.0030)∗∗

−0.0132
(0.0043)∗∗

−0.0133
(0.0031)∗∗

−0.0308
(0.0036)∗∗

−0.0005
(0.0027)

0.0166
(0.0233)

0.0050
(0.0168)

−0.0471
(0.0132)∗

−0.0517
(0.0099)∗∗

0.0266
(0.0023)∗∗

0.0082
(0.0020)∗∗

0.0033
(0.0020)

−0.0062
(0.0021)∗∗

0.0114
(0.0018)∗∗

0.0040
(0.0017)∗

0.0029
(0.0019)

−0.0058
(0.0019)

0.0766
(0.0043)∗∗

0.1092
(0.0052)∗∗

0.1293
(0.0062)∗∗

0.1493
(0.0076)∗∗

0.1554
(0.0088)∗∗

0.1548
(0.0099)∗∗

0.0463
(0.0035)∗∗

0.0680
(0.0042)∗∗

0.0816
(0.0051)∗∗

0.0992
(0.0062)∗∗

0.1076
(0.0073)∗∗

0.1157
(0.0083)∗∗

0.0472
(0.0161)∗∗

0.0450
(0.0141)∗∗

0.0736
(0.0052)∗∗

0.0958
(0.0053)∗∗

0.0994
(0.0052)∗∗

0.0978
(0.0053)∗∗

0.1119
(0.0054)∗∗

0.1073
(0.0061)∗∗

0.0467
(0.0039)∗∗

0.0647
(0.0039)∗∗

0.0743
(0.0038)∗∗

0.0796
(0.0039)∗∗

0.0886
(0.0039)∗∗

0.0913
(0.0044)∗∗

−0.0188
(0.0227)

−0.0144
(0.0180)

0.0365
(0.0170)∗

−0.0186
(0.0137)

−0.0382
(0.0050)∗∗

−0.0150
(0.0038)∗∗

−0.0049
(0.0025)∗
0.0356
(0.0143)∗

−0.0010
(0.0018)∗∗
0.0278
(0.0100)∗

0.0086
(0.0027)∗∗

−0.0004
(0.0020)

Charles T. Clotfelter, Helen F. muchacho, and Jacob L. Vigdor

Table A.4. Continuado.

Ecuación

Very competitive

Unranked

Mean teacher test score

Absences

OLS
MATH

3A.1

OLS
READING MATH

Maestros
Fixed Effects Fixed Effects

Maestros

READING

3A.2

3A.3

3A.4

0.0125
(0.0037)∗∗

0.0050
(0.0027)

−0.0120
(0.0074)

−0.0142
(0.0052)∗∗

0.0123
(0.0014)∗∗

0.0070
(0.0010)∗∗

Number of sick + personal days

−0.0024
(0.0001)∗∗

−0.0011
(0.0001)∗∗

−0.0017
(0.0001)∗∗

−0.0009
(0.0001)∗∗

Constant

0.6731
(0.0161)∗∗

0.5568
(0.0146)∗∗

0.6811
(0.0149)∗∗

0.5712
(0.0144)∗∗

Number of Observations

997,408

1,005,380

1,123,603

1,131,781

0.7200

0.6924

0.7551

0.7063

Notas: Standard errors are shown in parentheses. Dependent variable is normalized achievement
test score on North Carolina end-of-grade tests.
∗signiﬁcant at 5%; ∗∗signiﬁcant at 1%

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

mi
d
tu
mi
d
pag
a
r
t
i
C
mi
–
pag
d

F
/

4
2
1
1
5
1
6
8
9
1
1
9
mi
d
pag
2
0
0
9
4
2
1
1
5
pag
d