STATISTICAL
ANALYSIS
OF
SELENIUM
TOXICITY
DATA
­
PEER
REVIEW
DRAFTAugust
31,
1999
Prepared
for
U.
S.
Environmental
Protection
Agency
Health
and
Ecological
Criteria
Division
401
M
St.,
SW
Washington,
DC
20460
Contract
No.
68­
C7­
0002
Work
Assignment
No.
4­
31
Prepared
by
Jie
Tao,
Penelope
Kellar,
and
William
Warren­
Hicks
The
Cadmus
Group,
Inc.
1920
Highway
54
Executive
Park,
Suite
100
Durham,
NC
27713
DISCLAIMER
This
report
was
prepared
by
the
Cadmus
Group
Inc.
under
Contract
No.
68­
C7­
0002
to
the
U.
S.
Environmental
Protection
Agency.
The
information
in
this
document
does
not
necessarily
reflect
the
policy
of
the
U.
S.
Environmental
Protection
Agency
and
no
official
endorsement
should
be
inferred.
Mention
of
Trade
names
or
commercial
products
does
not
constitute
endorsement
or
recommendation
for
use.
iii
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
of
Contents
1.
BACKGROUND
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
2.
EXPERIMENTAL
DESIGN
AND
PROCEDURES
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2
3.
EFFECTS
OF
SELENIUM
ON
SURVIVAL
AND
GROWTH
OF
ADULT
BLUEGILLS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
3.1
SUMMARY
OF
SURVIVAL
AND
GROWTH
DATA
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
3.2
ANOVA
ON
PERCENT
SURVIVAL
AND
GROWTH
OF
ADULT
BLUEGILLS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
3.2.1
Survival
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
3.2.2
Growth
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
13
4.
EFFECTS
OF
SELENIUM
ON
SPAWNING
ACTIVITY
AND
PROGENY
OF
ADULT
BLUEGILLS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
16
4.1
SUMMARY
OF
FIELD
NEST
AND
EGG
CUP
DATA
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
16
4.2
ANOVA
ON
FIELD
NEST
AND
EGG
CUP
DATA
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
16
4.2.1
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
4.2.2
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
26
4.2.3
Study
III
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
4.2.4
Study
III
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
5.
SUMMARY
AND
CONCLUSIONS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
40
Appendix
A:
Calculations
of
Variables
in
Tables
3­
1,
3­
2,
4­
1,
and
4­
2
Appendix
B:
SAS
®
Output
From
PROC
MIXED
for
Field
Nest
and
Egg
Cup
Data
Analyses
Appendix
C:
Histograms
for
Survival
Data
Appendix
D:
Graphics
for
Field
Nest
and
Egg
Cup
Data
Appendix
E:
Data
Used
in
the
Analysis
Appendix
F:
SAS
®
Code
iv
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEList
of
Tables
2­
1.
Description
of
Datasets
in
the
MERS
Selenium
Studies
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
2­
2.
Selenium
Treatments
for
the
Three
Studies
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
2­
3.
Measured
Concentrations
of
Selenium
in
Water
During
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
3­
1.
Adult
Percent
Survival
and
Growth
During
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7
3­
2.
Adult
Percent
Survival
and
Growth
During
Study
III
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8
3­
3.
Adult
Growth
by
Gender
During
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
3­
4.
Adult
Growth
by
Gender
During
Study
III
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
10
3­
5.
ANOVA
Results
on
Percent
Survival:
Model
Significance
(
p­
Value)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
11
4­
1.
Spawning
Activity
and
Effects
on
Progeny
During
Study
II.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
17
4­
2.
Spawning
Activity
and
Effects
on
Progeny
During
Study
III.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
18
4­
3.
Summary
of
Average
Percent
Abnormalities
for
Study
II,
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
19
4­
4.
Summary
of
Average
Percent
Abnormalities
for
Study
II,
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
19
4­
5.
Summary
of
Average
Percent
Abnormalities
for
Study
III,
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
20
4­
6.
Summary
of
Average
Percent
Abnormalities
for
Study
III,
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
20
4­
7.
Experimental
Design
for
Field
Nest
Data
 
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
4­
8.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
22
4­
9.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
(
Arc­
sine
Square­
root
Transformed
Data)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
22
4­
10.
ANOVA
Results
From
PROC
MIXED:
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
4­
11.
Means
Test
(
Dunnett's)
From
PROC
MIXED:
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
24
4­
12.
Repeated­
measures
Design
for
Field
Nest
Data
 
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
4­
13.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
(
Repeated­
measures
Analysis)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
4­
14.
Repeated­
measures
ANOVA
Results
From
PROC
MIXED
 
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
26
4­
15.
Means
Test
(
Dunnett's)
for
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
26
4­
16.
Experimental
Design
for
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
28
4­
17.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
28
4­
18.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Arc­
sine
Square­
root
Transformed
Data)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
29
4­
19.
ANOVA
Results
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
29
4­
20.
Means
Test
(
Dunnett's)
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
29
4­
21.
Study
II
Egg
Cup
Data
 
Repeated­
measures
Design
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
30
4­
22.
Summary
of
Zero
Abnormalities
for
Study
II,
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
31
4­
23.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Repeated­
measures
Analysis
 
Full
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
4­
24.
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Full
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
4­
25.
Means
Test
(
Dunnett's)
for
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Full
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
4­
26.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Repeated­
measures
Analysis
 
Partial
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
34
v
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEList
of
Tables
(
continued)
 
4­
27.
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Partial
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
34
4­
28.
Means
Test
(
Dunnett's)
for
Repeated­
measures
ANOVA
From
PROC
MIXED
Study
II
Egg
Cup
Data
(
Partial
Time­
series
Dataset)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
4­
29.
Experimental
Design
for
Field
Nest
Data
 
Study
III
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
4­
30.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Field
Nest
Data
.
.
.
.
.
.
.
.
36
4­
31.
ANOVA
Results
From
PROC
MIXED
 
Study
III
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
36
4­
32.
Experimental
Design
for
Egg
Cup
Data
 
Study
III
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
38
4­
33.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
38
4­
34.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Egg
Cup
Data
(
Arc­
sine
Square­
root
Transformed
Data)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
39
4­
35.
ANOVA
Results
From
PROC
MIXED
 
Study
III
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
39
5­
1.
Summary
of
Statistical
Results
from
Study
II
Field
Nest
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
42
5­
2.
Summary
of
Statistical
Results
from
Study
II
Egg
Cup
Data
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
43
List
of
Figures
3­
1.
Residuals
from
the
ANOVA
Model
on
Percent
Survival
 
Study
II
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12
3­
2.
Residuals
from
the
ANOVA
Model
on
Percent
Survival
 
Study
III
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12
3­
3.
Power
of
the
ANOVA
Model
on
Percent
Survival
 
Study
II
Days
1­
221
.
.
.
.
.
.
.
.
.
.
.
14
3­
4.
Power
of
the
ANOVA
Model
on
Percent
Survival
 
Study
II
Days
1­
320
.
.
.
.
.
.
.
.
.
.
.
14
3­
5.
Power
of
the
ANOVA
Model
on
Percent
Survival
 
Study
III
Days
1­
181
.
.
.
.
.
.
.
.
.
.
.
15
3­
6.
Power
of
the
ANOVA
Model
on
Percent
Survival
 
Study
III
Days
1­
265
.
.
.
.
.
.
.
.
.
.
.
15
1
See
Hermanutz,
R.
O.,
K.
N.
Allen,
T.
H.
Rousch,
and
S.
Hedtke.
1992.
Effects
of
elevated
selenium
concentrations
on
bluegills
(
Lepomis
macrochirus)
in
outdoor
experimental
streams.
Environ.
Toxicol.
Chem.
11:
217­
224.

1
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESECTION
1
BACKGROUND
The
U.
S.
Environmental
Protection
Agency
(
EPA)
is
currently
in
the
process
of
revising
its
aquatic
life
criteria
for
selenium.
The
selenium
criteria
were
last
published
in
1987,
and
since
then,
additional
data
have
become
available
on
the
effects
of
selenium
on
aquatic
organisms.
Included
among
the
new
data
is
a
series
of
three
studies
on
the
effect
of
selenium
on
bluegills
(
Lepomis
macrochirus)
in
outdoor
experimental
streams
at
the
Monticello
Ecological
Research
Station
(
MERS)
in
Monticello,
Minnesota.
Results
from
the
first
of
the
three
studies
were
published
in
19921,
and
they
are
not
discussed
here.

In
this
report,
data
from
the
second
and
third
MERS
selenium
studies
(
Study
II
and
Study
III)
are
evaluated
by
The
Cadmus
Group,
Inc.,
using
a
variety
of
statistical
methods.
EPA
intends
to
consider
the
results
from
the
statistical
analyses
presented
in
this
report
in
its
forthcoming
manuscript
on
the
second
and
third
MERS
studies.
Section
2
of
this
report
provides
a
brief
summary
of
the
experimental
design
and
procedures
used
in
Studies
II
and
III.
Additional
detail
on
the
experimental
methodology
will
be
provided
by
EPA
in
its
forthcoming
manuscript.
Section
3
presents
the
statistical
analysis
of
the
effects
of
selenium
on
bluegill
spawning
and
progeny.
Section
4
presents
the
statistical
analysis
of
the
effects
of
selenium
on
bluegill
spawning
and
progeny.
Finally,
details
on
the
calculations,
statistical
output,
graphical
summaries,
raw
data,
and
the
statistical
programs
used
are
provided
in
Appendices
A
through
F.
2
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESECTION
2
EXPERIMENTAL
DESIGN
AND
PROCEDURES
This
section
presents
a
brief
summary
of
the
experimental
design
and
procedures
used
in
Studies
II
and
III
for
evaluating
the
effect
of
selenium
on
bluegills
in
outdoor
streams
at
the
Monticello
Experimental
Research
Station.
EPA
intends
to
provide
additional
detail
on
the
study
design
and
experimental
procedures
in
a
forthcoming
manuscript
on
Studies
II
and
III.
As
discussed
in
Section
1,
the
results
of
a
prior
study
(
Study
I)
are
not
analyzed
in
this
report.

Using
data
obtained
from
EPA,
The
Cadmus
Group,
Inc.,
(
Cadmus)
calculated
many
descriptive
statistics.
Some
data
were
corrected
for
transcription
errors
before
the
analysis.
EPA
provided
the
description
of
the
experimental
methodology
and
guidance
concerning
some
aspects
of
the
statistical
analyses,
such
as
the
use
of
PROC
MIXED
as
opposed
to
PROC
GLM.
During
the
course
of
data
evaluation,
EPA
and
Cadmus
considered
such
issues
as
the
following:
(
1)
methods
for
handling
missing
observations
and
data
inconsistencies,
(
2)
methods
for
handling
repeated
measures
over
time,
(
3)
data
transformations,
and
(
4)
issues
regarding
the
calculations.
The
datasets
used
in
our
analyses
are
listed
in
Table
2­
1.
Table
2­
2
provides
a
description
of
the
basic
experimental
layout
of
the
field
studies
conducted
at
the
Monticello
experimental
stream
sites.
Table
2­
3
presents
descriptive
statistics
on
the
concentrations
of
selenium
measured
during
each
study.

The
Monticello
Ecological
Research
Station
contains
eight
outdoor
streams
that
were
supplied
with
Mississippi
River
water
or
a
mixture
of
river
water
and
well
water.
Six
of
the
eight
streams
were
used
in
Study
I,
with
nominal
selenium
concentrations
of
0
µ
g/
L,
10
µ
g/
L,
and
30
µ
g/
L.
All
eight
streams
were
used
in
Studies
II
and
III.
For
Study
II,
dosing
of
the
10
µ
g/
L
selenium
streams
was
continued.
Dosing
of
the
30
µ
g/
L
streams
was
discontinued,
and
these
streams
were
used
to
determine
whether
residual
toxicity
was
caused
by
previous
selenium
dosing.
The
other
four
streams
(
2
unused,
2
controls
in
Study
I)
were
randomly
assigned
to
the
nominal
concentrations
of
0
µ
g/
L
and
2.5
µ
g/
L.
In
Study
III,
selenium
was
not
added
to
any
stream.
Study
III
addresses
only
residual
effects
in
recovering
streams.

For
both
studies,
adult
bluegills
were
obtained
from
a
south­
central
Minnesota
farm
pond.
Eightyfive
fish
in
Study
II
and
98
fish
in
Study
III
were
randomly
distributed
without
regard
to
sex
in
the
upper
reaches
(
i.
e.,
sampling
Stations
0­
2)
in
each
stream.
In
each
stream,
a
random
sample
of
the
adults
was
transferred
to
the
lower
reach
of
the
respective
experimental
stream
(
Station
6)
for
the
reproduction
portion
of
the
studies;
26
per
stream
were
transferred
in
Study
II
and
22
to
50
per
stream
in
Study
III.
Each
fish
was
weighed
at
the
time
of
transfer,
and
two
to
four
fish
were
randomly
selected
from
each
stream
for
measurement
of
selenium
in
selected
tissues.
After
transfer,
each
Station
6
pool
was
checked
daily
for
the
presence
of
bluegill
nests.
All
observed
nests
were
sampled
three
times
a
week
for
the
presence
of
embryos
and
larvae.
If,
after
five
passes
of
the
sampling
device,
no
embryos
or
larvae
were
present,
the
nest
was
considered
inactive.
In
Study
III,
the
sampling
device
was
passed
over
the
nest
until
an
adequate
sample
was
collected
or
the
investigator
determined
the
nest
to
be
inactive.
The
numbers
of
live
and
dead
embryos
and
larvae
were
recorded
(
hereinafter
referred
to
as
"
Field
Nest
Data").
3
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
2­
1.
Description
of
Datasets
in
the
MERS
Selenium
Studiesa
Dataset
File
Name
Date
Received
Description
Comments
Concerning
the
Datasets
Study
II
Survival
Data
s2surviv.
sd2
October
1997
number
of
fish
at
the
beginning
and
end
of
the
two­
phase
study
(
survival
and
growth,
reproduction)
for
each
stream
Growth
Data
bgmay89.
sd2
bgag8889.
sd2
October
1997
weight
and
gender
for
each
fish
on
the
transfer
date;
weight,
length,
K
factor,
and
gender
for
each
fish
at
the
end
of
study
Field
Nest
Data
fldabnrm.
sd2
October
1997,
April
1998
number
of
live
and
dead
embryos
and
larvae
in
each
nest,
age
of
the
larvae,
number
of
larvae
in
each
subsample,
and
number
of
larvae
with
abnormalities
in
the
subsample
Data
received
in
October
1997
were
replaced
by
the
corrected
version
received
in
April
1998;
duplicate
spawning
activities
in
a
given
nest
were
treated
as
independent
events.

Egg
Cup
Data
cupdays.
sd2
October
1997
number
of
live
and
dead
embryos
and
larvae
each
day
in
each
cup
taken
from
randomly
selected
nests,
and
number
of
larvae
with
abnormalities
each
day
in
each
cup
Discovered
some
measurement
errors;
for
example,
total
number
of
larvae
are
not
consistent
throughout
five­
day
experiment;
as
a
result,
the
calculated
percent
hatch
exceeds
100%
for
two
cups;
in
this
case,
we
truncated
the
values
to
100%
only
for
the
arc­
sine
square­
root
transformation.

Study
III
Survival
Data
s3surviv.
sd2
October
1997
number
of
fish
at
the
beginning
and
end
of
the
two­
phase
study
(
survival
and
growth,
reproduction)
for
each
stream
Growth
Data
bgmay90.
sd2
bgf8990.
sd2
October
1997
weight
and
gender
for
each
fish
on
the
transfer
date;
weight,
length,
and
gender
for
each
fish
at
the
end
of
study
Field
Nest
Data
s3nest.
sd2
October
1997
spawn
number,
number
of
larvae
in
each
subsample,
age
of
the
larvae
(
mostly
missing
and
number
of
larvae
with
abnormalities
in
the
subsample
Egg
Cup
Data
s3eggcup.
sd2
October
1997
initial
number
in
each
cup,
number
of
live
and
dead
embryos
and
larvae
each
day
in
each
cup
taken
from
randomly
selected
nests,
and
number
of
larvae
with
abnormalities
each
day
in
each
cup
Discovered
some
measurement
errors;
for
example,
the
initial
number
in
each
cup
is
sometimes
not
consistent
with
the
calculated
total
number
for
that
cup
at
Day
1;
in
this
case,
the
calculated
number
was
used
in
the
analysis;
total
number
of
larvae
is
not
consistent
throughout
the
five­
day
experiment;
as
a
result,
the
calculated
percent
hatch
exceeds
100%
for
one
cup,
and
in
this
case,
we
truncated
the
value
to
100%
only
for
the
arc­
sine
square­
root
transformation.

a
The
calculations
in
Tables
3­
1,
3­
2,
4­
1,
and
4­
2
are
summarized
in
Appendix
A.
The
complete
dataset
printouts
are
presented
in
Appendix
E.
4
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
2­
2.
Selenium
Treatments
for
the
Three
Studies
Addition
of
Selenium
Treatment
Streams
Total
Period
of
Dosing
Study
30
µ
g/
L
4
and
6
03/
11/
87
­
10/
02/
88
(
571
days)
I
10
µ
g/
L
3
and
8
03/
11/
87
­
11/
07/
89
(
972
days)
I,
II
2.5
µ
g/
L
2
and
7
10/
3/
88
­
11/
07/
89
(
400
days)
II
Studies
of
Bluegill
Stream
Study
Ia
Study
II
Study
III
Beginningb
Transferc
Endd
09/
01/
87
05/
16/
88
08/
22/
88
10/
88
05/
89
08/
89
11/
89
05/
90
07/
90
1
Unused
Control
Control
2
Unused
2.5
µ
g/
L
Recovering
3
10
µ
g/
L
10
µ
g/
L
Recovering
4
30
µ
g/
L
Recovering
Recovering
5
Control
Control
Control
6
30
µ
g/
L
Recovering
Recovering
7
Control
2.5
µ
g/
L
Recovering
8
10
µ
g/
L
10
µ
g/
L
Recovering
a
Previously
reported
(
see
Footnote
1,
page
1),
data
are
not
evaluated
in
this
report.
b
Adult
bluegills
placed
in
Stations
0­
2
for
survival
and
growth
study.
c
Transfer
of
adult
bluegills
within
each
stream
from
Stations
0­
2
to
Station
6
for
reproduction
study.
d
Adult
bluegills
removed
from
Station
6.

Table
2­
3.
Measured
Concentrations
of
Selenium
in
Water
During
Study
II
Intended
Concentration
2.5
µ
g/
L
10
µ
g/
L
Stream
Number
2
7
3
8
Stations
1
and
3
combined
Mean
2.48
2.67
8.87
9.65
Standard
Deviation
0.28
0.45
1.46
1.81
Number
of
Measurements
29
30
27
30
Stations
5
and
7
combined
Mean
2.67
2.55
10.40
10.83
Standard
Deviation
0.43
0.39
1.15
1.07
Number
of
Measurements
28
27
27
27
Mean
exposure
concentrationa
2.53
2.63
9.34
10.02
a
Mean
exposure
concentration
of
adult
bluegills
was
calculated
as
the
time­
weighted
mean
of
the
means
of
Stations
1
and
3
(
221
days)
and
Stations
5
and
7
(
99
days).
5
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESamples
of
embryos
were
incubated
in
the
laboratory
to
determine
hatchability,
larval
survival,
and
incidence
of
larval
anomalies.
Randomly
selected
embryos
were
reared
for
several
days
in
incubation
cups
(
hereinafter
referred
to
as
"
Egg
Cup
data").
Cup
contents
were
removed
and
examined
daily.
Percent
hatch
and
larval
survival
were
recorded.
Live
larvae
were
examined
for
abnormalities.

The
statistical
analyses
conducted
on
the
Monticello
field
and
laboratory
data
are
described
in
the
following
sections.
2
Montgomery,
D.
C.
1991.
"
Design
and
Analysis
of
Experiments."
Third
Edition.
John
Wiley
&
Sons.
pp.
73­
80.

6
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESECTION
3
EFFECTS
OF
SELENIUM
ON
SURVIVAL
AND
GROWTH
OF
ADULT
BLUEGILLS
This
section
presents
the
summary
and
analysis
approaches
for
the
survival
and
growth
data
for
Studies
II
and
III.
Section
3.1
summarizes
the
survival
and
growth
data.
The
analysis
of
variance
(
ANOVA)
models
applied
to
the
survival
and
growth
data
are
presented
in
Section
3.2
3.1
SUMMARY
OF
SURVIVAL
AND
GROWTH
DATA
Table
3­
1
summarizes
the
adult
bluegill
percent
survival
and
growth
data
during
Study
II
at
the
time
of
transfer
to
Station
6
(
for
the
reproductive
portion
of
the
experiment)
(
Day
221)
and
at
the
end
of
the
study
(
Day
320).
Table
3­
2
presents
a
similar
summary
for
Study
III
at
the
time
of
transfer
(
Day
181)
and
at
the
end
of
the
study
(
Day
265).
An
explanation
of
the
calculation
of
the
variables
in
the
tables
is
presented
in
Tables
A­
1
and
A­
2
in
Appendix
A.
To
evaluate
differences
in
growth
between
genders,
Cadmus
also
summarized
length,
weight,
and
K
factor
(
weight
×
10/
length3)
were
summarized
by
gender,
as
presented
in
Tables
3­
3
and
3­
4.

Throughout
this
report,
a
standardized
process
was
used
to
examine
the
results
of
an
ANOVA
model.
First,
an
ANOVA
model
that
is
consistent
with
the
experimental
design
was
selected.
The
model
was
fit
to
the
data.
An
F­
statistic
evaluating
the
relative
fit
of
the
model
to
the
data
was
examined.
If
the
p­
value
of
the
model
F­
statistic
was
less
than
0.05
(
p<
0.05),
the
model
was
considered
to
be
significant.
For
those
models
that
were
significant,
individual
parameters
in
the
model
were
examined.
Those
parameters
where
the
p­
value
of
the
parameter
F­
statistic
was
less
than
0.05
were
considered
significant
and
subjected
to
a
means
separation
test.
Several
means
tests
are
available
in
the
literature
and
most
software
packages.
Each
method
has
a
slightly
different
interpretation,
and
the
methods
are
not
guaranteed
to
produce
the
same
results.
Dunnett's
test
against
control
and
the
Tukey's
standardized
range
test
were
used
for
this
analysis.
2
Those
cases
in
which
the
two
tests
provide
different
results
are
noted
in
the
narrative
of
the
report.

3.2
ANOVA
ON
PERCENT
SURVIVAL
AND
GROWTH
OF
ADULT
BLUEGILLS
3.2.1
Survival
Percent
survival
at
the
transfer
date
and
at
the
end
of
the
study
was
analyzed
using
the
following
oneway
ANOVA
model
to
evaluate
the
effects
of
different
selenium
concentrations
(
0,
2.5,
and
10
µ
g/
L)
on
adult
bluegills.
The
30
µ
g/
L
treatment
was
excluded
from
this
ANOVA
for
Study
II,
because
this
treatment
represents
the
recovery
from
the
previous
study
and,
therefore,
is
not
comparable
to
the
continuous
exposure
regimes
of
the
2.5
and
10
µ
g/
L
treatments.
Similarly,
the
30
µ
g/
L
treatment
was
not
comparable
in
Study
III.
7
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
3­
1.
Adult
Percent
Survival
and
Growth
During
Study
IIa
Stream
Selenium
Treatment
Control
2.5
µ
g/
L
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Day
221b
%
survival
47.1
49.4
32.9
42.4
34.1
43.5
34.1
51.8
Weight
(
g)
c
103.0
(
25.4)
98.9
(
30.2)
101.8
(
34.6)
101.3
(
28.4)
100.0
(
32.8)
92.2
(
30.4)
98.1
(
36.5)
101.5
(
26.0)

Weight
gain
(
g)
d
24.8
20.7
23.6
23.1
21.8
14.0
19.9
23.3
Day
320e
Cumulative
%
survivalf
23.5
28.5
8.9
39.1
9.2
16.7
0
19.9
%
survival
from
Day
221
to
Day
320b,
e
50.0
57.7
26.9
92.3
26.9
38.5
0
38.5
Weight
(
g)
c
113.3
(
24.3)
141.3
(
26.1)
145.5
(
24.9)
146.1
(
34.2)
164.1
(
37.0)
130.8
(
19.5)
­
i
156.7
(
30.8)

Weight
gain
(
g)
g
35.1
63.1
67.3
67.9
85.9
52.6
­
i
78.5
Length
(
mm)
c
178.6
(
11.4)
191.9
(
12.4)
190.4
(
8.5)
190.3
(
10.4)
191.3
(
15.4)
182.5
(
8.9)
­
i
189.4
(
12.9)

K
factorc,
h
1.964
(
0.178)
1.984
(
0.121)
2.089
(
0.144)
2.080
(
0.179)
2.314
(
0.123)
2.142
(
0.143)
­
i
2.284
(
0.163)

a
The
initial
averages
(
and
standard
deviations)
were:
weight
=
78.2
(
26.4)
g,
length
=
164.5
(
15.6)
mm,
and
K
factor
=
1.711
(
0.355).
b
A
subset
of
fish
was
transferred
to
Station
6
on
Day
221
for
the
reproductive
portion
of
the
study.
c
Numbers
in
parentheses
are
standard
deviations.
d
Weight
gain
from
the
beginning
of
the
study,
calculated
by
subtracting
the
average
initial
weight
from
the
weight
on
Day
221.
e
The
study
ended
on
Day
320.
f
The
product
of
percent
survival
on
Day
221
and
percent
survival
from
Day
221
to
Day
320.
g
Weight
gain
from
the
beginning
of
the
study,
calculated
by
subtracting
the
average
initial
weight
from
the
weight
on
Day
320.
h
K
factor
=
weight
x
10
/
length3.
Numbers
in
parentheses
are
standard
deviations.
i
All
fish
died
by
this
time
in
Stream
4.
8
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
3­
2.
Adult
Percent
Survival
and
Growth
During
Study
IIIa
Stream
Selenium
Treatment
Control
Recovering
2.5
µ
g/
L
Recovering
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Day
181b
%
survival
33.7
34.7
53.1
27.6
32.7
24.5
33.7
28.6
Weight
(
g)
c
113.6
(
22.7)
115.1
(
20.9)
102.6
(
22.5)
111.4
(
30.7)
114.9
(
21.3)
102.1
(
17.8)
107.3
(
20.8)
112.9
(
23.6)

Weight
gain
(
g)
d
26.6
27.5
15.0
23.8
27.3
14.5
19.7
25.3
%
survival
from
Day
181
to
Day
265b,
e
29.0
31.3
34.0
48.0
30.0
36.4
61.3
­
i
Day
265e
%
survivalf
9.8
10.8
18.0
13.2
9.8
8.9
20.6
­
i
Weight
(
g)
c
119.6
(
36.9)
156.2
(
24.1)
123.7
(
26.7)
142.9
(
34.8)
164.6
(
40.3)
162.6
(
16.8)
130.3
(
25.2)
­
i
Weight
gain
(
g)
g
32.0
68.6
36.1
55.3
77.0
75.0
42.7
­
i
Length
(
mm)
c
178.0
(
12.8)
191.2
(
8.8)
179.1
(
11.6)
184.8
(
15.7)
194.7
(
14.2)
173.5
(
32.7)
183.7
(
10.2)
­
i
K
factorc,
h
2.059
(
0.202)
2.216
(
0.112)
2.120
(
0.095)
2.232
(
0.166)
2.207
(
0.290)
4.161
(
3.878)
2.077
(
0.119)
­
i
a
The
initial
averages
(
and
standard
deviations)
were:
weight
=
87.6
(
21.3)
g,
length
=
167.5
(
11.0)
mm,
and
K
factor
=
1.841
(
0.273).
b
A
subset
of
fish
was
transferred
to
Station
6
on
Day
181
for
the
reproductive
portion
of
the
study.
c
Numbers
in
parentheses
are
standard
deviations.
d
Weight
gain
from
the
beginning
of
the
study,
calculated
by
subtracting
the
average
initial
weight
from
the
weight
on
Day
181.
e
The
study
ended
on
Day
265.
f
The
product
of
percent
survival
on
Day
181
and
percent
survival
from
Day
181
to
Day
265.
g
Weight
gain
from
the
beginning
of
the
study,
calculated
by
subtracting
the
average
initial
weight
from
the
weight
on
Day
265.
h
K
factor
=
weight
x
10
/
length3.
Numbers
in
parentheses
are
standard
deviations.
i
Stream
6
was
removed
from
this
study
after
Day
181.
9
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
3­
3.
Adult
Growth
by
Gender
During
Study
IIa
Stream
Selenium
Treatment
Control
2.5
µ
g/
L
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Day
221b
Weight
(
g)

Female
Indeterminat
e
Male
96.9
(
24.1)

81.4
(
27.1)

117.0
(
19.4)
98.9
(
27.1)

69.7
(
24.5)

110.7
(
29.7)
99.4
(
33.1)

71.3
(
13.4)

124.8
(
35.4)
106.8
(
22.3)

65.8
(
19.6)

107.5
(
31.4)
106.0
(
31.7)

60.0
(
15.0)

107.4
(
29.2)
93.4
(
32.3)

59.3
(
5.1)

96.7
(
28.8)
89.4
(
21.7)

67.0
(
7.7)

139.9
(
29.1)
100.1
(
22.6)

80.1
(
19.4)

113.5
(
27.7)

Day
320c
Weight
(
g)

Female
Male
109.1
(
23.6)

136.1
(
16.7)
125.3
(
14.6)

149.3
(
27.5)
128.8
(
18.7)

167.7
(
6.0)
126.1
(
30.1)

158.1
(
31.6)
173.3
(
29.7)

160.5
(
42.1)
132.0
(
22.6)

130.5
(
20.3)
­
d
152.4
(
38.5)

161.0
(
24.7)

Length
(
mm)

Female
Male
177.0
(
11.7)

187.5
(
2.1)
186.0
(
8.5)

194.8
(
13.4)
185.3
(
6.6)

197.3
(
5.0)
184.4
(
9.9)

193.9
(
9.2)
199.5
(
14.8)

188.0
(
15.9)
182.0
(
18.4)

182.6
(
7.4)
­
d
188.2
(
18.4)

190.6
(
5.4)

K
factor
Female
Male
1.946
(
0.180)

2.061
(
0.183)
1.945
(
0.122)

2.003
(
0.123)
2.017
(
0.141)

2.185
(
0.092)
1.977
(
0.161)

2.142
(
0.164)
2.177
(
0.112)

2.369
(
0.080)
2.200
(
0.288)

2.127
(
0.112)
­
d
2.255
(
0.133)

2.312
(
0.199)

a
The
initial
averages
(
and
standard
deviations)
were:
weight
=
78.2
(
26.4)
g,
length
=
164.5
(
15.6)
mm,
and
K
factor
=
1711
(
0.355).
Numbers
in
parentheses
are
standard
deviations.

b
The
fish
were
transferred
on
Day
221.

c
The
study
ended
on
Day
320.

d
All
fish
died
by
this
time
in
Stream
4.
10
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTE­
11
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
3­
4.
Adult
Growth
by
Gender
During
Study
IIIa
Stream
Selenium
Treatment
Control
Recovering
2.5
µ
g/
L
Recovering
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Day
181b
Weight
(
g)

Female
Indeterminate
Male
108.8
(
19.1)

126.3
(
27.6)
102.1
(
18.0)

129.6
(
13.1)
99.8
(
17.6)

57.7
(
5.5)

114.2
(
21.0)
107.2
(
22.8)

76.4
(
18.6)

129.6
(
27.2)
107.5
(
18.2)

131.1
(
19.0)
97.1
(
10.7)

110.6
(
24.1)
100.2
(
15.8)

115.0
(
23.1)
105.8
(
19.7)

130.8
(
24.1)

Day
265c
Weight
(
g)

Female
Male
104.9
(
25.3)

171.2
(
14.4)
136.3
(
32.1)

164.7
(
15.6)
94.9
(
8.3)

139.3
(
18.4)
113.4
(
12.5)

172.5
(
20.2)
150.9
(
41.9)

175.6
(
39.9)
156.2
(
14.9)

173.2
(
16.5)
111.8
(
10.3)

150.9
(
20.1)
­
d
Length
(
mm)

Female
Male
173.0
(
9.0)

195.5
(
6.4)
184.0
(
12.1)

194.3
(
5.4)
166.3
(
4.6)

186.0
(
7.4)
173.2
(
8.7)

196.3
(
12.1)
192.8
(
18.2)

196.2
(
12.3)
176.6
(
22.6)

168.3
(
51.4)
176.9
(
6.3)

191.3
(
8.1)
­
d
K
factor
Female
Male
1.993
(
0.177)

2.288
(
0.031)
2.159
(
0.076)

2.241
(
0.120)
2.059
(
0.095)

2.154
(
0.080)
2.181
(
0.108)

2.285
(
0.205)
2.102
(
0.419)

2.290
(
0.131)
3.111
(
1.546)

5.911
(
6.364)
2.017
(
0.103)

2.144
(
0.102)
­
d
a
All
Study
III
streams
were
recovering
from
prior
selenium
exposures
in
Study
II
(
2.5
and
10
µ
g/
L)
and
Study
I
(
30
µ
g/
L).

The
initial
averages
(
and
standard
deviations)
were:
weight
=
87.6
(
21.3)
g,
length
=
167.5
(
11.0)
mm,
and
K
factor
=
1.841
(
0.273).

Numbers
in
parentheses
are
standard
deviations.

b
A
subset
of
fish
was
transferred
to
Station
6
on
Day
181
for
the
reproductive
portion
of
this
study.

c
The
study
ended
on
Day
265.

d
Stream
6
was
removed
from
this
study
after
Day
181.
12
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEThe
models
used
for
statistical
analyses
were:

Study
II:
P­
221
ij,
P­
320
ij
=
µ
+
T
i
+
 ij
Study
III:
P­
181
ij,
P­
265
ij
=
µ
+
T
i
+
 ij
where,

P­
221
ij
=
percent
survival
from
Day
1
to
Day
221,
for
the
ith
treatment
and
jth
stream;

P­
320
ij
=
percent
survival
from
Day
1
to
Day
320;
calculated
as
the
product
of
the
percent
survival
from
Day
1
to
Day
221
and
percent
survival
from
Day
221
to
Day
320,
for
the
ith
treatment
and
jth
stream;

P­
181
ij
=
percent
survival
from
Day
1
to
Day
181,
for
the
ith
treatment
and
jth
stream;

P­
265
ij
=
percent
survival
from
Day
1
to
Day
265;
calculated
as
the
product
of
the
percent
survival
from
Day
1
to
Day
181
and
percent
survival
from
Day
181
to
Day
265;

µ
=
overall
mean;

T
i
=
treatment
effect,
i
=
1
to
3;
and
 ij
=
random
error.

For
the
purpose
of
hypothesis
testing,
the
error
term
is
assumed
to
follow
a
normal
distribution
with
independent
realizations
of
the
data.
Because
the
dataset
contains
only
six
observations
spanning
three
treatments,
the
assumption
of
normal
independent
errors
may
not
hold.
Figures
3­
1
and
3­
2
present
plots
of
the
model
residuals
against
treatment.
Examination
of
the
residual
plots
shows
that
the
residual
variance
is
not
consistent
for
each
treatment.
Particularly,
the
residual
variance
is
largest
for
the
2.5
µ
g/
L
selenium
treatment.
We
transformed
the
response
variable
(
percent
survival)
using
the
arc­
sine
square­
root
transformation
and
reran
the
ANOVA
model.
The
results
of
these
runs
are
presented
in
Table
3­
5.

Table
3­
5.
ANOVA
Results
on
Percent
Survival:
Model
Significance
(
p­
value)

Approach
Study
II
Study
III
Day
1­
221
Day
1­
320
Day
1­
181
Day
1­
265
Arc­
sine
Square­
root
of
the
Response
Variables
0.263
0.604
0.618
0.086
As
shown
in
Table
3­
5,
p­
values
of
the
ANOVA
model
are
greater
than
0.05,
which
indicates
no
significant
differences
in
percent
survival
at
 
=
0.05
with
varying
concentrations
of
selenium
for
either
Study
II
or
Study
III.
The
p­
value
is
defined
as
the
probability
of
observing
a
sample
outcome
more
contradictory
to
H
o
(
no
difference
in
response
variables
with
different
selenium
concentrations)
than
the
observed
sample
result.
The
smaller
the
p­
value,
the
heavier
the
weight
of
the
sample
evidence
against
H
o.
13
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEFigure
3­
1.
Residuals
from
the
ANOVA
Model
on
Percent
Survival
C
Study
II
Figure
3­
2.
Residuals
from
the
ANOVA
Model
on
Percent
Survival
C
Study
III
3
Steel,
R.
G.
D.,
and
J.
H.
Torrie.
1960.
Principles
and
Procedures
of
Statistics,
McGraw­
Hill,
New
York,
NY.

14
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEAlthough
the
residual
plots
show
some
violations
of
the
error
assumptions,
the
ANOVA
results
are
little
affected
by
these
violations.
The
largest
impact
on
the
ANOVA
results
is
probably
due
to
the
small
overall
sample
size
and
the
small
number
of
within­
treatment
replicates.

We
calculated
the
power
of
the
ANOVA
model
using
standard
statistical
methods
(
Steel
and
Torrie,
19603).
The
power
of
the
test
is
illustrated
in
Figures
3­
3
through
3­
6
for
selected
Type
I
error
rates.
A
key
issue
in
the
use
of
any
statistical
method,
including
ANOVA,
is
the
number
of
experimental
units
required
for
a
specified
decision
criterion.
In
the
ANOVA
models
where
percent
survival
is
the
response
variable,
the
experimental
unit
or
replicate
is
the
stream.
The
null
hypothesis
inherent
in
the
model
is
that
mean
percent
survival
is
the
same
among
all
selenium
treatments.
The
ANOVA
model
is
fit
to
several
datasets
with
six
streams
and
three
selenium
treatments.
Figures
3­
3
through
3­
6
indicate
the
number
of
streams
required
to
meet
specified
levels
of
power
and
Type
I
error
for
tests
of
hypotheses
on
fish
survival
(
note:
the
calculations
are
model
specific,
and
extrapolating
the
results
to
different
endpoints
or
ANOVA
models
may
not
be
appropriate).
Power
is
the
probability
of
detecting
a
real
difference
of
10%
among
treatment
means,
and
the
Type
I
error
(
0.05,
0.10,
0.20)
represents
the
probability
of
erroneously
detecting
a
treatment
difference
when
none
exists.
The
streams
are
assumed
to
be
equally
allocated
among
three
treatments.
It
is
desirable
to
have
a
high
chance
of
classifying
the
treatment
means
as
equal,
when
they
are
(
i.
e.,
high
power,
as
indicated
in
the
upper
regions
of
the
abscissa).
Examination
of
the
plots
shows
that
six
streams
provide
reasonably
good
power
for
all
study­
specific
datasets
(
power
ranges
between
0.6
and
1.0)

3.2.2
Growth
While
weight
gain
for
individual
fish
is
not
available
in
the
datasets,
we
calculated
weight
gain
during
the
period
between
transfer
and
the
end
of
the
study
for
each
gender
in
each
stream.
These
genderstream
specific
data
were
evaluated
using
the
following
model:

W
ijk
=
µ
+
T
i
+
G
j
+
(
TG)
ij
+
S
k(
i)
+
 ijk
where,

W
ijk
=
weight
gain
between
the
transfer
and
the
end
of
the
study,
i
=
1
to
3,
j=
1
to
2,
k
=
1
to
2;

µ
=
overall
mean;

T
i
=
treatment
effect,
i
=
1
to
3;

G
j
=
gender
effect,
j
=
1
to
2;

(
TG)
ij
=
interaction
between
the
treatment
and
gender,
i
=
1
to
3,
j
=
1
to
2;

S
k(
i)
=
stream
effect,
k
=
1
to
2,
nested
within
treatment,
considered
as
a
random
effect;
and
 ijk
=
random
error.

The
results
indicate
that
no
effect
is
significant
at
 
=
0.05
for
weight
gain
in
either
Study
II
or
Study
III.
15
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEAlpha
=
0.05
Alpha
=
0.10
Alpha
=
0.20
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Total
Number
of
Streams
0
2
4
6
8
10
12
14
16
18
20
Figure
3­
3.
Power
of
the
ANOVA
Model
on
Percent
Survival
C
Study
II
Days
1­
221
Alpha
=
0.05
Alpha
=
0.10
Alpha
=
0.20
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Total
Number
of
Streams
0
2
4
6
8
10
12
14
16
18
20
Figure
3­
4.
Power
of
the
ANOVA
Model
on
Percent
Survival
C
Study
II
Days
1­
320
16
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEAlpha
=
0.05
Alpha
=
0.10
Alpha
=
0.20
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Total
Number
of
Streams
0
2
4
6
8
10
12
14
16
18
20
Figure
3­
5.
Power
of
the
ANOVA
Model
on
Percent
Survival
C
Study
III
Days
1­
181
Alpha
=
0.05
Alpha
=
0.10
Alpha
=
0.20
0.89
0.90
0.91
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1.00
Total
Number
of
Streams
0
2
4
6
8
10
12
14
16
18
20
Figure
3­
6.
Power
of
the
ANOVA
Model
on
Percent
Survival
C
Study
III
Days
1­
265
17
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESECTION
4
EFFECTS
OF
SELENIUM
ON
SPAWNING
ACTIVITY
AND
PROGENY
OF
ADULT
BLUEGILLS
After
fish
were
transferred
to
Station
6
in
each
stream,
each
pool
was
checked
daily
for
the
presence
of
bluegill
nests.
All
observed
nests
were
marked
with
labeled
stakes
and
sampled
every
Monday,
Wednesday,
and
Friday
for
the
presence
of
embryos
and
larvae.
The
numbers
of
live
and
dead
embryos
and
larvae
were
recorded.
Dead
embryos
were
identified
by
their
opaqueness.
Samples
of
live
larvae
were
observed
under
a
dissecting
microscope
to
identify
morphological
anomalies.
This
set
of
data
is
referred
to
as
"
Field
Nest"
data
in
the
following
text.

Samples
of
embryos
were
incubated
in
the
laboratory
to
determine
embryo
hatchability,
larval
survival,
and
incidence
of
larval
anomalies.
Randomly
selected
embryos
were
reared
for
several
days
in
incubation
cups
in
both
Studies
II
and
III.
The
embryos
were
exposed
to
a
proportional
diluter
that
contained
the
same
river
water
and
the
same
nominal
selenium
concentrations
as
those
in
the
respective
test
streams.
Cup
contents
were
removed
and
examined
daily;
live
embryos
and
larvae
were
returned
after
the
cup
was
cleaned.
Percent
hatch
and
larval
survival
were
recorded.
Live
larvae
were
examined
for
abnormalities.
This
set
of
data
is
referred
to
as
"
Egg
Cup"
data
in
the
following
text.

The
Field
Nest
data
and
Egg
Cup
data
for
both
studies
are
summarized
in
Section
4.1.
Section
4.2
describes
the
experimental
design,
the
ANOVA
model,
and
results
for
these
datasets.

4.1
SUMMARY
OF
FIELD
NEST
AND
EGG
CUP
DATA
Tables
4­
1
and
4­
2
summarize
the
adult
bluegill
spawning
activity
and
effects
on
progeny
under
the
influence
of
differing
selenium
concentrations
for
Study
II
and
Study
III,
respectively.
The
number
of
active
nests,
number
of
embryos
collected,
number
of
larvae
collected,
and
number
of
samples
containing
larvae
were
summarized
for
each
stream
using
the
Field
Nest
Data.
In
addition,
the
percentage
of
dead
embryos,
percentage
of
dead
larvae,
and
percentage
of
abnormalities,
among
live
larvae
were
also
calculated
and
summarized
for
each
stream.
The
egg
cup
data
were
used
to
calculate
percent
hatch,
percent
survival
to
the
third
day
(
to
eliminate
the
starvation
effect
later
in
the
experiment),
percent
abnormalities
and
percent
healthy
among
live
larvae
for
each
stream.
Numbers
in
parentheses
are
the
standard
deviations
of
the
parameter.
Tables
A­
3
and
A­
4
in
Appendix
A
explain
the
calculations
for
each
variable
in
the
tables.

The
average
percent
abnormalities
among
live
larvae
on
each
day
at
each
selenium
concentration
for
Field
Nest
Data
and
egg
cup
data
for
Study
II
and
Study
III
are
summarized
and
presented
in
Tables
4­
3
through
4­
6.
In
addition,
a
series
of
plots
was
generated
to
depict
the
percent
abnormalities
among
live
larvae
on
different
days
at
different
dosing
levels
for
the
Field
Nest
Data
and
egg
cup
data
for
Studies
II
and
III.
These
figures
are
presented
in
Appendix
C.

4.2
ANOVA
ON
FIELD
NEST
AND
EGG
CUP
DATA
To
examine
the
effects
of
selenium
on
adult
bluegill
spawning
activity
and
progeny,
candidate
ANOVA
models
were
evaluated
and
those
models
most
consistent
with
the
experimental
design
were
selected.
Because
the
datasets
are
highly
unbalanced
and
random
effects
(
e.
g.,
stream,
nest)
are
present
in
the
models,
PROC
MIXED
provided
by
SAS
®
was
used
for
the
analyses
presented
in
this
section.
Detailed
information
concerning
PROC
MIXED
can
be
found
in
"
SAS/
STAT
Software:
4
Latour
D.,
K.
Latour,
and
R.
D.
Wolfinger.
1994.
Getting
started
with
PROC
MIXED.
SAS
Institute,
Inc.,
Cary,
NC.
5
Wolfinger
R.,
R.
Tobias,
and
J.
Sall.
1991.
Mixed
models:
A
future
direction.
Proceedings
of
the
Sixteenth
Annual
SAS
Users
Group
Conference,
SAS
Institute,
Inc.,
Cary,
NC.
pp.
1380­
1388.
6
Schwarz,
C.
J.
1993.
The
mixed­
model
ANOVA:
The
truth,
the
computer
packages,
the
books.
The
American
Statistician
47(
1):
48­
59.

18
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEChanges
and
Enhancement
through
Release
6.12."
Additional
information
on
PROC
MIXED
can
be
found
in
Latour
et
al.
(
1994),
4
Wolfinger
et
al.
(
1994),
5
and
Schwarz
(
1993).
6
Table
4­
1.
Spawning
Activity
and
Effects
on
Progeny
During
Study
II
Stream
Selenium
Treatment
Control
2.5
µ
g/
L
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Field
Nest
Data
#
of
active
nestsa
6
9
1
5
2
3
0
8
#
of
embryos
collecteda
2,458
1,329
0
1,462
672
931
­­­
646
%
dead
embryosa
0.94
0
­­­
0
0
0.32
­­­
0
#
larvae
collecteda
3,252
3,435
2,497
4,717
5,376
750
­­
6,788
%
dead
larvaea
0.03
1.05
0.20
0.08
0.50
0.40
­­­
7.79
#
of
samples
containing
larvae
(
n)
7
13
3
8
9
4
­­­
16
%
edemab
0
(
0­
0)
0
(
0­
3.3)
4.1
(
4.1­
4.1)
0
(
0­
10.7)
81.4
(
66.7­
96.2)
50.0
(
0­
91.2)
­­­
27.3
(
0­
91.3)

%
lordosisb
0
(
0­
0)
0
(
0­
0)
25.0
(
25.0­
25.0)
0
(
0­
3.2)
5.0
(
3.3­
6.7)
14.7
(
0­
23.3)
­­­
0
(
0­
6.7)

%
hemorrhagingb
0
(
0­
0)
0
(
0­
2.8)
77.6
(
77.6­
77.6)
52.0
(
0­
100.0)
55.5
(
23.1­
87.9)
26.7
(
20.6­
57.1)
­­­
17.1
(
0­
22.7)

Egg
Cup
Data
#
of
trialsc
6
5
0
4
3
2
­­­
6
%
hatchd
93.0
(
6.4)
96.4
(
3.1)
­­­
81.4
(
11.9)
83.3
(
23.1)
91.1
(
9.1)
­­­
92.9
(
12.7)

%
survival
to
third
daye
75.2
(
14.8)
71.5
(
22.1)
­­­
71.6
(
7.4)
57.7
(
32.1)
57.1
(
25.7)
­­­
79.0
(
14.2)

%
edemaf
0
(
0­
3.7)
0
(
0­
10.0)
­­­
0
(
0­
20.0)
100.0
(
0­
100.0)
100.0
(
74.1­
100)
­­­
17.4
(
0­
94.3)

%
lordosisf
0
(
0­
8.7)
0
(
0­
3.8)
­­­
0
(
0­
32.1)
11.1
(
0­
51.1)
18.2
(
0­
40.7)
­­­
0.0
(
0­
37.1)

%
hemorrhagingf
0
(
0­
4.3)
0
(
0­
10.0)
­­­
3.6
(
0­
81.4)
49.3
(
0­
100.0)
41.1
(
0­
83.3)
­­­
11.5
(
0­
45.7)

%
healthyg
97.8
(
91.3­
100)
97.9
(
90­
100)
­­­
92.2
(
18.6­
100)
0
(
0­
100)
0
(
0­
25.9)
­­­
70.7
(
5.7­
100)

a
Cumulative
for
the
stream,
i.
e.,
one
value
per
stream.
b
Among
live
larvae;
the
median
and
range
(
in
parentheses)
of
the
maximum
incidence
per
set
of
nest
observations
are
given.
c
A
trial
was
set
up
whenever
sufficient
larvae
were
collected.
d
Cumulative
percent
hatch
for
each
cup;
the
mean
and
standard
deviation
(
in
parentheses)
are
given.
e
Mean
percent
survival
to
third
day
after
first
larva
hatched;
the
standard
deviation
is
given
in
parentheses.
f
Among
larvae
that
survived
up
to
third
day
after
first
larva
hatched;
the
median
and
range
(
in
parentheses)
of
the
maximum
incidence
are
given.
g
Among
live
larvae
that
survived
up
to
third
day
after
first
larvae
hatched;
assumes
the
observations
of
multiple
abnormality
types
always
cooccurred
in
the
same
organism,
this
may
overestimate
the
actual
%
healthy
when
this
assumption
is
violated.
The
median
and
range
(
in
parentheses)
of
the
percent
healthy
per
cup
are
given.
19
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
2.
Spawning
Activity
and
Effects
on
Progeny
During
Study
IIIa
Stream
Selenium
Treatment
Control
Recovering
2.5
µ
g/
L
Recovering
10
µ
g/
L
Recovering
30
µ
g/
L
1
5
2
7
3
8
4
6
Field
Nest
Data
#
of
active
nestsb
4
3
4
2
3
4
7
­­­

#
of
samples
containing
larvae
(
n)
6
3
5
2
9
6
13
­­­

%
edemac
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
4.7)
­­­

%
lordosisc
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
2.3)
­­­

%
hemorrhagingc
0
(
0­
0)
0
(
0­
0)
2.44
(
0­
10.7)
0
(
0­
0)
0
(
0­
0)
0
(
0­
2.0)
2.38
(
0­
12.5)
­­­

Egg
Cup
Data
#
of
trialsd
2
3
3
7
­­­
3
5
­­­

%
hatche
85.3
(
3.8)
76.9
(
20.1)
90.0
(
6.0)
88.0
(
12.5)
­­­
78.9
(
13.3)
92.5
(
13.7)
­­­

%
survival
to
third
dayf
62.9
(
12.5)
68.0
(
19.0)
71.3
(
27.3)
72.2
(
11.2)
­­­
63.4
(
3.7)
81.1
(
21.2)
­­­

%
edemag
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
­­­
0
(
0­
0)
­­­
­­­

%
lordosisg
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
­­­
0
(
0­
0)
­­­
­­­

%
hemorrhagingg
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
0
(
0­
0)
­­­
0
(
0­
3.03)
­­­
­­­

%
healthyh
100
(
100­
100)
100
(
100­
100)
100
(
100­
100)
100
(
100­
100)
­­­
100
(
97­
100)
­­­
­­­

a
All
Study
III
streams
were
recovering
from
prior
selenium
exposures
in
Study
II
(
2.5
and
10
µ
g/
L)
or
Study
I
(
30
µ
g/
L).
b
Cumulative
for
the
stream,
i.
e.,
one
value
per
stream.
c
Among
live
larvae;
the
median
and
range
(
in
parentheses)
of
the
maximum
incidence
per
set
of
nest
observations
are
given.
d
A
trial
was
set
up
whenever
sufficient
larvae
were
collected.
e
Cumulative
percent
hatch
for
each
cup;
the
mean
and
standard
deviation
(
in
parentheses)
are
given.
f
Mean
percent
survival
to
third
day
after
first
larva
hatched;
the
standard
deviation
is
given
in
parentheses.
g
Among
larvae
that
survived
up
to
third
day
after
first
larva
hatched;
the
median
and
range
(
in
parentheses)
of
the
maximum
incidence
are
given.
h
Among
larvae
that
survived
up
to
third
day
after
first
larvae
hatched;
assumes
the
observations
of
multiple
abnormality
types
always
co­
occurred
in
the
same
organism,
this
may
overestimate
the
actual
%
healthy
when
this
assumption
is
violated.
The
median
and
range
(
in
parentheses)
of
the
minimum
percent
healthy
per
cup
are
given.
20
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
3.
Summary
of
Average
Percent
Abnormalities
for
Study
II,
Field
Nest
Data
Age
Selenium
Treatment
Control
2.50
µ
g/
L
10.00
µ
g/
L
Recovering
30.00
µ
g/
L
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.

1
...
...
...
...
1
0
0
11.1
3
24.4
11.1
8.9
1
0
0
0
2
4
0.8
0
0.7
3
1.4
2.4
59.2
6
35.1
0.5
35.5
1
0
0
22.7
3
6
0
0
0
4
0.4
0
26.0
2
78.9
7.4
41.8
5
25.5
0.9
12.4
4
2
0
0
0
1
10.7
0
14.3
...
...
...
...
2
9.3
3.3
5.0
5
2
0
0
0
1
0
25.0
0
...
...
...
...
...
...
...
...

6
5
0
0
0
...
...
...
...
...
...
...
...
1
72.7
0
13.6
7
1
0
0
0
...
...
...
...
...
...
...
...
1
80.0
2.9
17.1
Table
4­
4.
Summary
of
Average
Percent
Abnormalities
for
Study
II,
Egg
Cup
Data
Day
Selenium
Treatment
Control
2.50
µ
g/
L
10.00
µ
g/
L
Recovering
30.00
µ
g/
L
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthya
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthya
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthya
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthya
1
22
0
0.4
0.2
99.4
6
0
0
25.6
74.4
12
15.2
7.3
22.8
69.9
6
2.5
2.2
6.0
92.5
2
22
0.5
0.5
0.6
98.9
12
0.4
1.2
1.6
97.4
11
57.7
9.4
27.6
35.1
11
3.1
1.5
3.2
94.4
3
22
0.6
0.2
1.0
99.0
12
2.7
3.1
5.1
91.1
11
97.6
10.0
35.1
2.4
11
33.5
6.4
15.4
64.3
4
22
1.2
0.2
1.5
98.1
12
8.9
1.5
3.2
90.4
10
90.0
15.0
18.4
0.0
11
70.5
1.0
23.4
28.6
5
16
0
0.9
0.0
99.1
11
1.0
4.9
3.1
93.0
4
100.0
19.3
6.3
0.0
9
83.6
6.2
11.1
16.4
a
%
healthy
=
[#
of
live
larvae
­
max(#
of
edema,
#
of
lordosis,
#
of
hemor.)]/#
of
live
larvae

100.
21
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
5.
Summary
of
Average
Percent
Abnormalities
for
Study
III,
Field
Nest
Dataa
Age
Selenium
Treatment
Control
Recovering
2.50
µ
g/
L
Recovering
10.00
µ
g/
L
Recovering
30.00
µ
g/
L
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.
#
of
obs.
%
edema
%
lordosis
%
hemor.

0.5
...
...
...
...
1
0
0
0
1
0
0
0
...
...
...
...

1
...
...
...
...
1
0
0
10.7
...
...
...
...
...
...
...
...

1.5
...
...
...
...
...
...
...
...
...
...
...
1
0
0
0
2
...
...
...
1
0
0
0
...
...
...
...
4
1.2
0
4.1
3
1
0
0
0
...
...
...
...
...
...
...
...
1
0
0
0
3.5
...
...
...
...
1
0
0
0
...
...
...
...
...
...
...
...

4
...
...
...
...
...
...
...
...
...
...
...
...
1
0
0
0
4.5
...
...
...
...
1
0
0
0
...
...
...
...
...
...
...
...

a
All
Study
III
streams
were
recovering
from
prior
selenium
exposures
in
Study
II
(
2.5
and
10
µ
g/
L)
or
Study
I
(
30
µ
g/
L).

Table
4­
6.
Summary
of
Average
Percent
Abnormalities
for
Study
III,
Egg
Cup
Dataa
Day
Selenium
Treatment
Control
Recovering
2.50
µ
g/
L
Recovering
10.00
µ
g/
L
Recovering
30.00
µ
g/
L
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthyb
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthyb
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthyb
#
of
obs.
%
edema
%
lordosis
%
hemor.
%
healthyb
1
1
0
0
0
100.0
7
0
0
0
100.0
3
0
0
0.7
100.0
2
0
0
0
100.0
2
5
0
0
0
100.0
10
0
0
0
100.0
3
0
0
0
100.0
5
0
0
0
100.0
3
5
0
0
0
100.0
10
0
0
0
100.0
2
0
0
0
100.0
5
0
0
0
100.0
4
5
0
0.6
0.6
99.4
9
0
0
0
100.0
3
0
0
0
100.0
5
0
0
0
100.0
5
5
0
0
0
100.0
10
0
0.4
0
99.6
3
0
0
0
100.0
5
0
0
0
100.0
a
All
Study
III
streams
were
recovering
from
prior
selenium
exposures
in
Study
II
(
2.5
and
10
µ
g/
L)
or
Study
I
(
30
µ
g/
L).

b
%
healthy
=
[#
of
live
larvae
­
max
(#
of
edema,
#
of
lordosis,
#
of
hemor.)]/#
of
live
larvae

100.
22
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEAs
a
part
of
the
analysis,
tests
for
normality
and
homogeneity
of
variance
were
performed
on
the
appropriate
data.
As
appropriate,
data
transformations
were
applied
to
adjust
for
non­
normal
model
errors.

Two
types
of
models
were
fit
to
the
data.
First,
we
fit
a
standard
ANOVA
model
that
does
not
account
for
time­
based
effects
associated
with
the
repeated
sampling
of
nests
or
egg
cups
during
the
course
of
the
experiment.
The
response
variables
were
the
maximum
incidences
(
e.
g.,
maximum
%
edema,
maximum
%
hemorrhaging)
per
nest
or
cup.
For
egg
cup
data,
the
minimum
%
healthy
was
also
included
as
one
of
the
response
variables.
Note
that
the
value
of
%
healthy
was
calculated
for
the
live
larvae
only,
and
under
the
assumption
that
observations
of
multiple
abnormality
types
always
occurred
in
combination
on
the
same
group
of
"
affected"
organisms.
Therefore,
this
may
result
in
an
overestimation
of
the
actual
%
healthy
when
such
an
assumption
is
violated.
Second,
we
fit
a
form
of
the
ANOVA
model
associated
with
repeated­
measures
analysis.
This
model
adjusts
the
model
results
for
any
time­
dependent
correlations
that
exist
among
the
data.
Each
model
is
described
below.

In
addition,
for
Study
II,
none
of
the
information
associated
with
30
µ
g/
L
selenium
treatments
was
included
in
the
analytical
dataset
to
which
the
ANOVAs
were
fit.
We
excluded
this
treatment
because
it
represents
the
recovery
from
the
previous
study
and
therefore
is
not
comparable
to
the
continuous
exposure
regimes
of
the
2.5
and
10
µ
g/
L
treatments.

4.2.1
Study
II
Field
Nest
Data
Experimental
Design
The
experimental
design
for
the
Study
II
Field
Nest
data,
ignoring
time­
series
effects,
is
summarized
in
Table
4­
7.

Table
4­
7.
Experimental
Design
for
Field
Nest
Data
 
Study
II
Selenium
Treatment
Stream
Maximum
%
Abnormalities
from
Each
Spawna
Control
1
C2,
C3,
D1,
D2,
D3,
D4
5
B1,
B11,
B3,
B4,
B5,
B7,
C2,
C4
2.50
µ
g/
L
2
D1
7
B6,
C2,
C3,
C4,
C8
10.0
µ
g/
L
3
A11,
A12,
A21,
A22
8
B1,
B3,
C1
Recovering
30.0
µ
g/
L
4
No
Data
6
A1,
A12,
A13­
1,
A13­
2,
A2­
1,
A2­
2,
A3,
A4,
A5,
A8
a
Table
entries
(
e.
g.,
C2)
represent
the
spawns
that
have
the
values
for
the
response
variables.

In
this
formulation,
a
single
value
for
each
spawn
in
each
nest
is
the
data
evaluated
by
the
ANOVA.
The
table
indicates
the
nest
number
for
each
spawn.
Where
multiple
spawns
occurred
in
a
nest,
the
spawns
are
indexed.
Note
that
the
recovering
30
µ
g/
L
treatment
was
not
evaluated
in
the
ANOVA.
Only
the
maximum
percent
abnormalities
from
each
spawn
are
used
in
the
analysis.

ANOVA
Analysis
The
following
mixed
model
was
performed
on
the
Field
Nest
Data,
with
stream
considered
as
the
random
effect:
23
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTER
ijk
=
µ
+
C
i
+
S
j(
i)
+
 ijk
where,

R
ijk
=
multivariate
response:
%
edema,
%
lordosis,
and
%
hemorrhaging
(
all
maximum
incidence
per
nest);

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;
and
 ijk
=
random
error.

A
test
of
normality
on
the
model
residuals
(
 ijk)
and
a
homogeneity
of
variance
test
among
treatment
levels
were
performed.
An
arc­
sine
square­
root
transformation
was
applied
to
the
data
when
model
assumptions
of
normality
and/
or
homogeneity
of
variance
were
violated.
In
cases
where
an
arc­
sine
square­
root
transformation
was
inadequate,
a
ranking
transformation
was
substituted.
The
results
are
summarized
in
Table
4­
8.

Table
4­
8.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
Normality
Homogeneity
of
Variance
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

p­
value
0.0001*
0.0001*
0.0103*
<
0.0001*
**
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.
**
Non­
calculable,
given
the
values
in
the
dataset.

Table
4­
8
indicates
that
the
data
generally
do
not
satisfy
the
normality
and
homogeneity
of
variance
assumptions
and
thus
require
transformation.
Therefore,
an
arc­
sine
square­
root
transformation
was
applied
to
the
data.
The
results
of
repeated
testing
are
summarized
in
Table
4­
9.

Table
4­
9.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
(
Arc­
sine
Square­
root
Transformed
Data)

Normality
Homogeneity
of
Variance
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

p­
value
0.0004*
0.0013*
0.0032*
<
0.0001*
**
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.
**
Non­
calculable,
given
the
values
in
the
dataset.

Again,
the
data
did
not
pass
the
tests.
Because
both
the
raw
data
and
the
transformed
data
failed
the
normality
and
homogeneity
of
variance
tests,
the
ANOVA
was
also
performed
on
ranks
(
Iman,
7
Iman,
R.
L.
1982.
Some
Aspects
of
the
Rank
Transform
in
Analysis
of
Variance
Problems.
Seventh
Annual
SAS
Users
Group
International
Conference.
8
Self,
S.
G.,
and
K.
Y.
Liang.
1987.
Asymptotic
properties
of
maximum
likelihood
estimators
and
likelihood
ratio
tests
under
nonstandard
conditions.
J.
Amer.
Stat.
Assoc.
82:
605­
610.

24
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTE­
19827).
Here,
we
ranked
the
response
variables
from
lowest
to
highest
across
treatments
and
reran
the
ANOVA.
Fitting
an
ANOVA
model
to
ranks
may
not
be
appropriate.
First,
the
model
residuals
are
non­
normal.
Second,
the
residuals
are,
by
definition,
truncated
within
the
range
of
the
lowest
and
highest
rank.
Therefore,
applying
standard
hypothesis
testing
techniques,
which
required
the
assumption
of
normally
distributed
residuals
with
zero
mean,
may
not
be
appropriate.
Generally,
analysis
of
ranked
data
proceeds
with
nonparametric
approaches,
thereby
negating
the
distributional
assumptions
inherent
in
parametric
techniques.
However,
because
nonparametric
methods
are
not
available
for
mixed­
model
ANOVA
designs
(
the
model
that
is
most
consistent
with
the
experimental
design
of
this
study),
the
results
of
the
ANOVA
were
presented
on
both
raw
and
ranked
data.
The
results
of
performing
the
ANOVA
on
both
the
raw
data
and
the
ranks
are
summarized
in
Table
4­
10.

Table
4­
10.
ANOVA
Results
From
PROC
MIXED
 
Study
II
Field
Nest
Data
Parameter
Test
of
Parameter
Significance
(
p­
value,
raw
data)
Test
of
Parameter
Significance
(
p­
value,
ranks)

Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

Treatment
0.0233*
0.5510
0.0568
0.0271*
0.1907
0.0129*

Stream(
treatment),
random
**
0.2669
**
**
0.6313
**

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zero.

The
missing
p­
values
in
Table
4­
10
and
in
the
following
tables
from
PROC
MIXED
result
from
the
fact
that
the
variance
component
estimates
are
zero,
and
therefore
test
statistics
cannot
be
computed.
The
zero
variance
estimates
may
arise
for
several
reasons.
For
example,
the
variability
in
the
data
may
be
large
enough
to
produce
a
negative
estimate
(
negative
values
were
restricted
to
zeros),
even
though
the
true
value
of
the
variance
component
is
positive;
data
may
contain
outliers
and
a
different
model
for
interpreting
the
data
may
be
appropriate.
Although
alternative
models
might
better
address
the
missing
p­
value
issue,
such
alternative
models
were
not
used,
because
they
were
judged
to
be
incompatible
with
the
experimental
design.

Note
that
the
test
on
the
random
effect,
stream(
treatment),
provided
by
PROC
MIXED
is
based
on
large
sample
asymptotic
theory8
and
therefore
may
not
be
appropriate
in
this
application.
The
random
effect
results
should
therefore
be
interpreted
with
caution.
Table
4­
10
indicates
that
ANOVAs
on
the
raw
and
ranked
data
provide
generally
consistent
results
with
respect
to
determination
of
significant
parameters.
Selenium
concentration
appears
to
be
an
important
component
of
the
maximum
percent
incidence
determinations
for
%
edema
and
perhaps
%
hemorrhaging,
but
not
for
%
lordosis.
To
examine
which
treatments
differ,
the
means
test
from
PROC
MIXED,
which
adjusts
the
standard
error
of
the
treatment
for
the
random
effects
in
the
model,
was
performed.
The
results
are
presented
in
Table
4­
11.
Dunnett's
multiple
comparison
against
control
was
performed
on
this
dataset
and
all
subsequent
datasets.

The
results
indicate
that,
for
maximum
%
edema,
the
selenium
treatment
mean
of
10
µ
g/
L
differed
from
the
control
mean,
using
both
the
raw
data
and
the
rank
transformed
data.
For
maximum
%
25
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEhemorrhaging
the
selenium
treatment
mean
of
2.5
µ
g/
L
differed
from
the
control
mean
for
the
rank
transformed
data
(
treatment
effects
were
marginally
not
significant
at
2.5
µ
g/
L
for
the
raw
data).
In
addition,
the
mean
of
10
µ
g/
L
differed
from
the
control
mean
for
ranked
data
only
for
maximum
%
hemorrhaging.

Table
4­
11.
Means
Test
(
Dunnett's)
From
PROC
MIXED
 
Study
II
Field
Nest
Data
Selenium
Treatments
Test
of
the
Difference
(
p­
value,
raw
data)
Test
of
the
Difference
(
p­
value,
ranks)

Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

2.5
­
0.0
µ
g/
L
0.962
0.478
0.050*,
a
0.358
0.285
0.022*

10.0
­
0.0
µ
g/
L
0.020*
0.687
0.133
0.020*
0.166
0.014*

*
Significant
p­
value
at
 
=
0.05.
a
Treatment
effects
were
marginally
not
significant
at
 
=
0.05,
see
Table
4­
10.

Repeated­
Measures
ANOVA
For
the
Field
Nest
Data,
each
nest
was
sampled
two
or
three
times
a
week
using
a
plastic
tube
to
determine
the
presence
of
embryos
and
larvae.
Table
4­
12
presents
the
experimental
design
for
the
Field
Nest
Data
using
time­
based
samples.
If
the
nest
is
considered
the
sampling
unit,
then
a
repeated­
measures
analysis
is
appropriate.
The
table
shows
that
the
number
of
times
a
nest
is
sampled
is
not
consistent
over
the
experiment.
Due
to
a
large
number
of
missing
values
in
the
dataset
for
ages
other
than
two
and
three
days,
only
these
data
were
retained
in
the
dataset
for
the
repeated­
measures
analysis.
The
repeated­
measures
ANOVA
model
corresponding
to
the
experimental
design
can
be
written
as
follows:

R
ijklm
=
µ
+
C
i
+
S
j(
i)
+
N
k(
j)
+
A
l
+
(
CA)
il
+
 ijklm
where,

R
ijklm
=
multivariate
response:
%
edema,
%
lordosis,
and
%
hemorrhaging;

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;

N
k(
j)
=
spawn
effect,
nested
within
stream,
considered
as
a
random
effect,
and
the
subject
for
the
repeated­
measures
analysis;

A
l
=
age
effect,
k
=
1
to
7
days;

(
CA)
il
=
interaction
between
treatment
and
age,
i
=
1
to
3,
k
=
1
to
7;
and
 ijklm
=
random
error.

Before
the
analysis,
normality
and
homogeneity
of
variance
were
tested.
The
results
are
summarized
in
Table
4­
13.
Because
the
raw
data
failed
the
test
for
normality,
the
arc­
sine
square­
root
transformation
was
applied
to
the
data,
and
tests
of
normality
and
homogeneity
of
variance
were
then
performed
on
the
transformed
data.
These
results
are
also
included
in
Table
4­
13.
26
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
12.
Repeated­
measures
Design
for
Field
Nest
Data
 
Study
II
Selenium
Treatment
Stream
Nest
%
Abnormalities
at
Age
(
Days)
1
2
3
4
5
6
7
Control
1
D1
x
x
D2
x
x
D3
x
x
D4
x
5
B1
x
B11
x
x
B3
x
x
B4
x
x
B5
x
x
B7
x
C3
x
C4
x
x
2.5
µ
g/
L
2
D1
x
x
x
7
B6
x
C3
x
x
x
C4
x
x
C8
x
10
µ
g/
L
3
A1­
1
x
x
A1­
2
x
x
A2­
1
x
x
A2­
2
x
8
B1
x
B3
x
C1
x
x
Recovering
30
µ
g/
L
4
No
data
6
A12
x
A13­
1
x
x
A13­
2
x
A2­
1
x
A2­
2
x
A3
x
x
A4
x
A5
x
x
A8
x
Table
4­
13.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Field
Nest
Data
(
Repeated­
measures
Analysis)

p­
value
using
Normality
Homogeneity
of
Variance
%
edema
%
lordosis
%
hemor.
%
edema
%
lordosis
%
hemor.

Raw
data
0.0001*
0.0001*
0.0570
<
0.0001*
**
<
0.0001*

Arc­
sine
transformed
data
0.0005*
0.0001*
0.0478*
<
0.0001*
**
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
all
zero
values
of
%
lordosis
for
the
control.
As
a
result,
the
standard
deviation
cannot
be
computed,
and
the
test
cannot
be
performed.
27
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEBecause
both
the
raw
data
and
the
transformed
data
failed
the
normality
test,
the
ANOVA
was
performed
on
ranked
data
as
well.
The
ranks
were
constructed
independently
for
ages
of
2
and
3
days.
The
results
for
the
ANOVAs
are
presented
in
Table
4­
14.

Table
4­
14.
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Field
Nest
Data
Parameters
p­
value
(
raw
data)
p­
value
(
ranks)

%
edema
%
lordosis
%
hemor.
%
edema
%
lordosis
%
hemor.

Treatment
0.0294*
0.1518
0.1211
0.2023
0.1788
0.0443*

Age
0.1617
0.2038
0.5031
0.4264
0.6817
0.5881
Age
×
Treatment
0.1389
0.0140*
0.4758
0.3318
0.0215*
0.4965
Stream(
treatment),
random
**
0.8353
**
0.5790
**
**

*
Significant
p­
value
at
 
=
0.05.

**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zeroes.

Table
4­
14
indicates
that
the
interaction
between
age
and
treatment
is
significant
for
%
lordosis,
using
both
the
raw
data
and
rank
transformed
data;
or,
in
other
words,
selenium
effects
differ
for
different
ages.
As
a
result,
the
means
test
on
the
age
×
treatment
interaction
for
%
lordosis
should
be
examined.
The
results
("
Tests
of
Effect
Slices"
in
SAS
®
output
and
Table
5­
1)
show
that
10
µ
g/
L
selenium
had
a
significant
effect
on
%
lordosis
at
age
3
(
raw
data).
At
2.5
µ
g/
L,
significant
effects
on
%
lordosis
also
occurred
at
age
2
(
ranked
data).
Regarding
treatment
effects,
repeated
measures
analysis
indicates
that
selenium
had
a
significant
effect
at
10
µ
g/
L
on
%
edema
(
raw
data)
and
%
hemorrhaging
(
ranked
data).
Age
does
not
appear
to
have
a
significant
effect
on
any
of
the
three
measures.
The
complete
output
from
PROC
MIXED
is
presented
in
Appendix
B.
The
results
of
the
means
test
(
Dunnett's)
for
the
treatment
effect
from
PROC
MIXED
are
summarized
in
Table
4­
15.

Table
4­
15.
Means
Test
(
Dunnett's)
for
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Field
Nest
Data
Selenium
Treatment
Test
of
the
difference
(
p­
value,
raw
data)
Test
of
the
difference
(
p­
value,
ranks)

%
edema
%
lordosis
%
hemor.
%
edema
%
lordosis
%
hemor.

2.5
­
0.0
µ
g/
L
0.999
0.638a
0.121
0.681
0.167a
0.075
10.0
­
0.0
µ
g/
L
0.027*
0.118a
0.163
0.159
0.252a
0.041*

*
Significant
p­
value
at
 
=
0.05.
a
Significant
p­
value
was
found
at
 
=
0.05
for
treatment
×
interaction
(
see
Table
4­
14).

4.2.2
Study
II
Egg
Cup
Data
ANOVA
Analysis
The
experimental
design
for
Study
II
Egg
Cup
data,
ignoring
time
effects,
is
summarized
in
Table
4­
16.
Again,
the
information
associated
with
the
recovering
30
µ
g/
L
selenium
treatment
was
not
used
in
the
analysis.
Only
incidence
rates
to
the
third
day
of
sampling
were
evaluated.
The
data
analysis
was
truncated
at
three
days
because
the
starvation
effect
will
more
likely
be
confounded
with
the
9
%
healthy
=
[#
of
live
larvae
­
max(#
of
edema,
#
of
lordosis,
#
of
hemor.)]
/#
of
live
larvae
*
100
for
each
observation.
This
calculation
assumes
the
observations
of
multiple
abnormality
types
always
co­
occurred
in
the
same
group
of
"
affected"
organisms;
hence
the
value
may
be
overestimated
when
a
violation
of
the
assumption
occurs.

28
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEselenium
effect
after
the
third
day
of
the
experiment.
Based
on
the
experimental
design,
the
following
ANOVA
model
was
implemented:

R
ijkl
=
µ
+
C
i
+
S
j(
i)
+
N
k(
j)
+
 ijkl
where,

R
ijkl
=
%
hatch,
%
survival
(
to
the
third
day),
%
edema
(
maximum
incidence
to
the
third
day),
%
lordosis
(
maximum
incidence
to
the
third
day),
%
hemorrhaging
(
maximum
incidence
to
the
third
day),
%
healthy9
(
minimum
incidence
to
the
third
day);

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;

N
k(
j)
=
spawn
effect,
nested
within
stream,
considered
as
a
random
effect;
and
 ijkl
=
random
error.

Tests
for
normality
on
model
residuals
and
homogeneity
of
variance
among
treatments
on
the
response
variables
were
performed
on
the
dataset.
The
results
are
summarized
in
Table
4­
17.

Because
the
data
failed
the
normality
test
for
most
of
the
response
variables,
and
failed
the
homogeneity
of
variance
test
for
all
the
variables,
the
arc­
sine
square­
root
transformation
was
applied
to
the
data,
and
tests
of
normality
on
model
residuals
and
homogeneity
of
variance
on
the
transformed
data
were
performed.
The
results
are
summarized
in
Table
4­
18.

This
transformation
resulted
in
an
improved
normality
test
on
model
residuals
and
homogeneity
of
variance
test
for
only
one
out
of
six
response
variables,
i.
e.,
maximum
%
hemorrhaging
and
maximum
%
edema,
respectively.
The
dataset
was
then
rank
transformed,
and
the
ANOVA
was
performed
on
both
the
raw
data
and
the
rank
transformed
data.
The
results
are
summarized
in
Table
4­
19.

Examination
of
Table
4­
19
shows
a
treatment
effect
for
maximum
%
edema
and
minimum
%
healthy
using
both
raw
and
ranked
data.
Means
testing
on
these
parameters
show
a
significant
difference
between
the
mean
control
effect
and
the
mean
10
µ
g/
L
selenium
effect.
The
p­
values
of
the
means
test
are
provided
in
Table
4­
20.

Repeated­
measures
ANOVA
For
the
egg
cup
data,
each
cup
was
sampled
for
five
days
to
observe
the
abnormalities
among
live
larvae.
To
account
for
any
time­
dependent
correlation,
a
repeated­
measures
ANOVA
was
performed.
The
experimental
design
for
the
repeated­
measures
analysis
is
summarized
in
Table
4­
21.
Before
performing
any
analysis,
the
egg
cup
data
were
examined
for
the
occurrence
of
zero
incidence
29
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEvalues
as
summarized
in
Table
4­
22.
Accompanying
figures
showing
the
zero
percent
abnormalities
are
also
presented
in
Appendix
D.
The
analysis
is
performed
on
two
datasets:
the
full
time­
series
dataset
and
the
partial
time­
series
dataset.

Table
4­
16.
Experimental
Design
for
Study
II
Egg
Cup
Data
Selenium
Treatment
Stream
Spawn
(
nest)
%
Survival
to
the
Third
Day,
Maximum
%
Abnormalities
to
the
Third
Daya
Control
1
1
(
D1)
S1,
S2
2
(
D2)
S1,
S2
3
(
D3)
S1,
S2
4
(
D4)
S1,
S2
5
(
C2)
S1,
S2
6
(
C3)
S1,
S2
5
1
(
B3)
S1,
S2
2
(
B7)
S1,
S2
3
(
C2)
S1,
S2
4
(
C3)
S1,
S2
5
(
C4)
S1,
S2
2.5
µ
g/
L
2
No
Data
7
1
(
C3)
S1,
S2,
S3,
S4,
S5,
S6
3
(
C4)
S1,
S2
4
(
C2)
S1,
S2
5
(
C8)
S1,
S2
10
µ
g/
L
3
1
(
A1)
S1,
S2,
S3,
S4
2
(
A1)
S1,
S2
(
only
1
day's
data
available)

3
(
A2)
S1,
S2
8
1
(
B1)
S1,
S2
2
(
B3)
S1,
S2
Recovering
30
µ
g/
L
4
No
Data
6
1
(
A4)
S1,
S2
2
(
A3)
S1
3
(
A8)
S1,
S2
4
(
A11)
S1,
S2
5
(
A13)
S1,
S2
6
(
A13)
S1,
S2
a
Sx:
Multivariate
response
observed
for
sample
x
(
or
cup
x)

Table
4­
17.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
Normality
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Min
%
healthy
p­
value
0.2096
0.8156
0.0001*
0.0001*
0.0008*
0.0001*
30
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEHomogeneity
of
Variance
p­
value
<
0.0001*
0.0004*
<
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.

Table
4­
18.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Arc­
sine
Square­
root
Transformed
Data)

Normality
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Min
%
healthy
p­
value
0.9239
0.8939
0.0001*
0.0001*
0.2270
0.0040*

Homogeneity
of
Variance
p­
value
0.0015*
0.0004*
0.0638
0.0011*
<
0.0001*
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.

Table
4­
19.
ANOVA
Results
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
Factors
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Min
%
healthy
p­
value
(
raw
data)

Treatment
0.2770
0.2629
0.0012*
0.0597
0.0841
0.0129*

Stream(
treat),
random
**
**
**
**
**
**

Spawn(
stream),
random
0.0118*
0.0305*
0.0993
**
0.0068*
0.0148*

p­
value
(
rank
transformed
data)

Treatment
0.4061
0.3852
0.0362*
0.0548
0.0647
0.0491*

Stream(
treat),
random
**
**
**
**
**
**

Spawn(
stream),
random
0.0200*
0.0502
0.0514
**
0.0289*
0.0344*

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zeroes.

Table
4­
20.
Means
Test
(
Dunnett's)
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
Selenium
Treatment
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Min
%
healthy
Test
of
the
difference
(
p­
value,
raw
data)

2.5
­
0.0
µ
g/
L
0.504
0.982
0.562
0.609
0.297
0.132
10.0
­
0.0
µ
g/
L
0.248
0.230
0.001*
0.050*
0.070
0.010*

Test
of
the
difference
(
p­
value,
ranks)

2.5
­
0.0
µ
g/
L
0.417
0.890
0.547
0.508
0.195
0.223
Table
4­
20.
Means
Test
(
Dunnett's)
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
Selenium
Treatment
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Min
%
healthy
31
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTE­
10.0
­
0.0
µ
g/
L
0.501
0.332
0.030*
0.045*,
a
0.055
0.040*

*
Significant
p­
value
at
 
=
0.05.
a
Treatment
effects
were
marginally
nonsignificant
at
 
=
0.05
(
see
Table
4­
19).

Table
4­
21.
Study
II
Egg
Cup
Data
 
Repeated­
measures
Design
Selenium
Treatment
Stream
Spawn
(
nest)
Cup
(
or
Sample)
Multivariate
Response
on
Day
1
2
3
4
5
Control
1
1
(
D1)
1
2
x
x
x
x
x
x
x
x
x
x
2
(
D2)
1
2
x
x
x
x
x
x
x
x
x
x
3
(
D3)
1
2
x
x
x
x
x
x
x
x
x
x
4
(
D4)
1
2
x
x
x
x
x
x
x
x
x
x
5
(
C2)
1
2
x
x
x
x
x
x
x
x
x
x
6
(
C3)
1
2
x
x
x
x
x
x
x
x
x
x
5
1
(
B3)
1
2
x
x
x
x
x
x
x
x
x
x
2
(
B7)
1
2
x
x
x
x
x
x
x
x
x
3
(
C2)
1
2
x
x
x
x
x
x
x
x
4
(
C3)
1
2
x
x
x
x
x
x
x
x
x
5
(
C4)
1
2
x
x
x
x
x
x
x
x
2.5
µ
g/
L
2
No
Data
7
1
(
C3)
1
2
3
4
5
6
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
3
(
C4)
1
2
x
x
x
x
x
x
x
x
x
x
Table
4­
21.
Study
II
Egg
Cup
Data
 
Repeated­
measures
Design
Selenium
Treatment
Stream
Spawn
(
nest)
Cup
(
or
Sample)
Multivariate
Response
on
Day
1
2
3
4
5
32
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTE­
4
(
C2)
1
2
x
x
x
x
x
x
x
x
x
5
(
C8)
1
2
x
x
x
x
x
x
x
x
Table
4­
21.
Study
II
Egg
Cup
Data
 
Repeated­
measures
Design
Selenium
Treatment
Stream
Spawn
(
nest)
Cup
(
or
Sample)
Multivariate
Response
on
Day
1
2
3
4
5
33
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTE­
10
µ
g/
L
3
1
(
A1)
1
2
3
4
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
2
(
A1)
1
2
x
x
x
x
x
3
(
A2)
1
2
x
x
x
x
x
x
x
x
8
1
(
B1)
1
2
x
x
x
x
x
x
x
x
x
x
2
(
B3)
1
2
x
x
x
x
x
x
x
Recovering
30
µ
g/
L
4
No
Data
6
1
(
A4)
1
2
x
x
x
x
x
x
x
x
x
x
2
(
A3)
1
No
Data
x
x
x
x
3
(
A8)
1
2
x
x
x
x
x
x
x
x
x
4
(
A11)
1
2
x
x
x
x
x
x
x
x
5
(
A13)
1
2
x
x
x
x
x
x
x
x
x
6
(
A13)
1
2
x
x
x
x
x
x
x
x
Table
4­
22.
Summary
of
Zero
Abnormalities
for
Study
II,
Egg
Cup
Data
Day
Total
#
of
Obs.
Total
#
(
Percent)
of
Zero
Abnormalities
Edema
Hemorrhaging
Lordosis
1
46
41
(
89.1%
)
34
(
73.9%
)
40
(
87.0%
)

2
56
37
(
66.1%
)
35
(
62.5%
)
37
(
66.1%
)

3
56
32
(
57.1%
)
28
(
50.0%
)
41
(
73.2%
)

4
55
30
(
54.5%
)
28
(
50.9%
)
43
(
78.2%
)

5
40
26
(
65.0%
)
28
(
70.0%
)
32
(
80.0%
)
34
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEFull
time­
series
dataset
Due
to
the
relatively
high
incidence
of
zero
abnormalities
at
Day
1
for
%
edema
and
%
lordosis
(
see
Table
4­
22),
these
observations
were
eliminated
from
the
repeated­
measures
ANOVA.
The
resulting
dataset
is
referred
to
as
a
"
full
time­
series"
dataset.
The
following
model
based
on
the
experimental
design
presented
in
Table
4­
21
was
applied
to
this
dataset:

R
ijklmn
=
µ
+
C
i
+
S
j(
i)
+
N
k(
j)
+
Cp
l(
k)
+
A
m
+
(
CA)
im
+
 ijklmn
where,

R
ijklmn
=
multivariate
response:
%
edema,
%
lordosis,
%
hemorrhaging,
and
%
healthy;

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;

N
k(
j)
=
spawn
effect,
nested
within
streams,
considered
as
a
random
effect;

Cp
l(
k)
=
cup
effect,
nested
within
spawns,
considered
as
a
random
effect;

A
m
=
time
effect,
m
=
1
to
5
days
for
%
hemorrhaging
and
%
healthy,
m
=
1
to
4
days
for
%
edema
and
%
lordosis;

(
CA)
im
=
interaction
between
treatment
and
time,
i
=
1
to
3,
k
=
1
to
5
(
or
1
to
4);
and
 ijklmn
=
random
error.

Again,
tests
for
normality
on
model
residuals
and
homogeneity
of
variance
on
the
response
variables
were
performed
on
this
dataset.
The
results
are
summarized
in
Table
4­
23.
Because
the
data
generally
failed
the
normality
test
for
%
edema,
%
lordosis,
and
%
hemorrhaging,
the
arc­
sine
squareroot
transformation
was
applied
to
the
data,
and
tests
of
normality
on
model
residuals
and
homogeneity
of
variance
on
the
response
variables
were
performed
on
the
transformed
data.
The
results
are
also
summarized
in
Table
4­
23.

Because
both
the
raw
data
and
the
transformed
data
failed
the
normality
and
the
homogeneity
of
variance
tests,
the
ANOVA
was
also
performed
on
the
ranked
response
variables.
The
results
are
summarized
in
Table
4­
24.

Examination
of
Table
4­
24
shows
that
the
interaction
between
age
(
day)
and
treatment
for
the
full
time­
series
data
sets
is
significant
for
%
edema
(
raw
data),
%
hemorrhaging
(
raw
and
ranked),
and
%
healthy
(
raw
and
ranked),
indicating
that
selenium
effects
differ
for
different
ages.
As
a
result,
means
tests
were
conducted
on
the
day
x
treatment
interaction
for
these
variables.
The
results
("
Tests
of
Effect
Slices"
in
SAS
®
output
and
Table
5­
2)
show
that
10
µ
g/
L
selenium
had
a
significant
effect
on
%
edema
(
days
2­
5,
raw),
%
hemorrhaging
(
days
1­
3,
raw;
days
1­
4,
ranked),
and
%
healthy
(
days
1­
5,
raw
and
ranked).
At
2.5
µ
g/
L,
selenium
had
a
significant
effect
on
%
hemorrhaging
(
day
1,
raw)
and
%
healthy
(
day
1,
raw).
Detailed
information
can
be
found
in
Appendix
B
"
PROC
MIXED
Output
from
SAS
®
.
"
For
measures
where
no
significant
interaction
between
age
and
35
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
23.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Repeated­
measures
Analysis
Full
Time­
series
Dataset)

p­
value
using
Normality
Homogeneity
of
Variance
%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
Raw
data
0.0001*
0.0001*
0.0001*
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*

Arc­
sine
square­
root
transformed
data
0.0001*
0.0001*
0.0001*
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.

Table
4­
24.
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Full
Time­
series
Dataset)

Factor
p­
value
(
raw
data)
p­
value
(
rank
transformed
data)

%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
Treatment
0.0559
0.2974
0.0510
0.0080*
0.0375*
0.0391*
0.0303*
0.0130*

Day
0.0001*
0.3980
0.0174*
0.0001*
0.0001*
0.0041*
0.0041*
0.0002*

Day
×
Treatment
0.0001*
0.7361
0.0064*
0.0001*
0.0804
0.0856
0.0104*
0.0302*

Stream(
treat),
random
0.3653
0.3770
**
**
0.7730
**
0.8831
**

Spawn(
stream),
random
0.1691
**
0.1920
0.0238*
0.0515
**
0.5898
0.3136
*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zeroes.

treatment
was
found,
repeated
measures
analysis
and
subsequent
means
testing
indicate
that
selenium
had
a
significant
effect
at
10
µ
g/
L
on
%
edema
(
ranked)
and
%
lordosis
(
ranked).
Table
4­
25
contains
the
complete
results
of
the
means
testing.
Age
(
day)
effects
were
also
significant
on
%
edema
and
%
lordosis
for
the
ranked
data.

Table
4­
25.
Means
Test
(
Dunnett's)
for
Repeated­
Measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Full
Time­
Series
Dataset)

Selenium
Treatment
Test
of
the
Difference
(
p­
value,
raw
data)
Test
of
the
Difference
(
p­
value,
ranks)

%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
2.5
­
0.0
µ
g/
L
0.977a
0.952
0.256a
0.211a
0.684
0.269
0.171a
0.122a
10.0
­
0.0
µ
g/
L
0.051a
0.264
0.042*,
a
0.007*,
a
0.032*
0.032*
0.025*,
a
0.011*,
a
*
Significant
p­
value
at
 
=
0.05.
a
Significant
p­
value
was
found
at
 
=
0.05
for
treatment
x
day
interaction
(
see
Table
4­
24).

Partial
time­
series
dataset
To
address
the
concern
that,
during
the
last
days
of
the
experiment,
a
starvation
effect
could
affect
fish
response
and
thus
confound
the
selenium
effect,
a
second
ANOVA
model
was
implemented.
36
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEThis
model,
termed
the
partial
time­
series
repeated­
measures
ANOVA,
was
performed
on
the
egg
cup
data
after
eliminating
observations
for
Days
4
and
5
from
the
full
time­
series
dataset.
The
form
of
the
ANOVA
model
is
the
same
as
that
described
above.

The
tests
for
normality
on
model
residuals
and
the
homogeneity
of
variance
among
treatments
on
the
response
variables
for
this
partial
time­
series
dataset
were
performed.
The
results
are
summarized
in
Table
4­
26.
Because
the
data
generally
failed
the
normality
and
homogeneity
of
variance
tests,
an
arc­
sine
square­
root
transformation
was
applied
to
the
data,
and
tests
of
normality
and
homogeneity
of
variance
were
performed
on
the
transformed
data.
The
results
are
also
summarized
in
Table
4­
26.

Table
4­
26.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
II
Egg
Cup
Data
(
Repeated­
measures
Analysis
 
Partial
Time­
series
Dataset)

p­
value
using
Normality
Homogeneity
of
Variance
%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
Raw
data
0.0001*
0.0001*
0.0001*
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*

Arc­
sine
square­
root
transformed
data
0.0001*
0.0001*
0.0001*
0.0008*
<
0.0001*
<
0.0001*
<
0.0001*
<
0.0001*

*
Significant
p­
value
at
 
=
0.05.

Because
both
the
raw
data
and
the
transformed
data
failed
the
normality
and
the
homogeneity
of
variance
tests,
the
ANOVA
was
also
performed
on
the
ranked
response
variables.
The
results
are
summarized
in
Table
4­
27.

Table
4­
27.
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Partial
Time­
series
Dataset)

Factor
p­
value
(
raw
data)
p­
value
(
rank
transformed
data)

%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
Treatment
0.0945
0.0618
0.0580
0.0227*
0.0201*
0.1427
0.0384*
0.0244*

Day
0.0001*
0.6258
0.2804
0.0001*
0.7070
0.8787
0.0736
0.6089
Day
×
Treatment
0.0001*
0.7862
0.0259*
0.0001*
0.4508
0.4639
0.0071*
0.0154*

Stream(
treat),
random
0.3504
**
**
**
**
0.4679
**
**

Spawn(
stream),
random
0.8744
**
0.1218
0.0189*
0.1198
**
0.0600
0.1102
*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zeroes.

Examination
of
Table
4­
27
shows
that
the
interaction
between
age
(
day)
and
treatment
for
the
partial
time­
series
data
sets
is
significant
for
%
edema
(
raw
data),
%
hemorrhaging
(
raw
and
ranked),
and
%
healthy
(
raw
and
ranked),
indicating
that
selenium
effects
differ
for
different
ages.
These
results
are
consistent
with
those
of
the
full
time­
series
dataset
(
Table
4­
25).
Means
tests
were
conducted
on
the
day
x
treatment
interaction
for
these
variables.
The
results
("
Tests
of
Effect
Slices"
in
SAS
®
37
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEoutput
and
Table
5­
2)
show
that
10
µ
g/
L
selenium
had
a
significant
effect
on
%
edema
(
days
2
and
3,
raw),
%
hemorrhaging
(
days
1­
3,
raw
and
ranked),
and
%
healthy
(
days
1­
3,
raw
and
ranked).
At
2.5
µ
g/
L,
selenium
had
a
significant
effect
on
%
hemorrhaging
(
day
1,
raw)
and
%
healthy
(
day
1,
raw).
Detailed
information
can
be
found
in
Appendix
B
"
PROC
MIXED
Output
from
SAS
®
.
"
For
measures
where
no
significant
interaction
between
age
and
treatment
was
found,
repeated
measures
analysis
and
subsequent
means
testing
indicate
that
selenium
had
a
significant
effect
at
10
µ
g/
L
on
%
edema
(
ranked)
and
%
lordosis
(
ranked).
Table
4­
28
contains
the
complete
results
of
the
means
testing.
No
statistically
significant
age
(
day)
effects
were
found
for
measures
shown
to
have
no
significant
age
and
treatment
interaction.

Table
4­
28.
Means
Test
(
Dunnett's)
for
Repeated­
measures
ANOVA
From
PROC
MIXED
 
Study
II
Egg
Cup
Data
(
Partial
Time­
series
Dataset)

Selenium
Treatment
Test
of
the
Difference
(
p­
value,
raw
data)
Test
of
the
Difference
(
p­
value,
ranks)

%
edema
%
lordosis
%
hemor.
%
healthy
%
edema
%
lordosis
%
hemor.
%
healthy
2.5
­
0.0
0.998a
0.563
0.252a
0.300a
0.723
0.918
0.259a
0.235a
10.0
­
0.0
0.089a
0.051
0.048*,
a
0.018*,
a
0.017*
0.126
0.031*,
a
0.020*,
a
*
Significant
p­
value
at
 
=
0.05.
a
Significant
p­
value
was
found
at
 =
0.05
for
treatment
×
day
interaction
(
see
Table
4­
27).

4.2.3
Study
III
Field
Nest
Data
ANOVA
Analysis
The
experimental
design
for
the
Study
III
Field
Nest
Data,
ignoring
time
effects,
is
summarized
in
Table
4­
29.
As
before,
observations
associated
with
the
30
µ
g/
L
selenium
treatment
were
not
evaluated.

Table
4­
29.
Experimental
Design
for
Field
Nest
Data
 
Study
III
Selenium
Treatment
Stream
Maximum
%
Abnormalities
from
Each
Spawn
Control
1
1,
2,
3,
4
5
1,
2,
3
2.5
µ
g/
L
2
1,
2,
3,
4
7
1,
2
10.0
µ
g/
L
3
1,
2,
3
8
1,
2,
3,
4
30.0
µ
g/
L
4
1,
2,
3,
4,
5,
6,
7
6
No
Data
The
following
mixed
model
was
performed
on
the
Study
III
Field
Nest
Data,
with
stream
considered
as
the
random
effect:

R
ijk
=
µ
+
C
i
+
S
j(
i)
+
 ijk
where,
38
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTER
ijk
=
multivariate
response:
%
edema,
%
lordosis,
and
%
hemorrhaging
(
all
maximum
incidence
per
nest);

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;
and
 ijk
=
random
error.

As
before,
the
normality
of
the
model
residuals
and
the
homogeneity
of
variance
among
treatment
levels
for
the
dependent
variables
were
tested.
The
results
are
summarized
in
Table
4­
30.
The
arcsine
square­
root
transformation
was
also
applied
to
the
dataset,
and
the
tests
for
normality
and
homogeneity
of
variance
on
the
transformed
data
were
performed
and
also
summarized
in
Table
4­
30.

Table
4­
30.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Field
Nest
Data
p­
value
using
Normality
Homogeneity
of
Variance
Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

Raw
data
**
**
0.0001*
**
**
***

Arc­
sine
square­
root
transformed
data
**
**
0.0001*
**
**
***

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
are
due
to
all
zero
values
in
the
dataset.
***
Not
calculable,
given
the
values
in
the
dataset.

The
unavailability
of
test
results
(
missing
p­
values)
for
maximum
%
edema
and
%
lordosis
results
from
the
fact
that
the
values
for
these
variables
are
all
zeroes
for
treatments
0,
2.5,
and
10
µ
g/
L
in
the
dataset.
Because
both
the
raw
data
and
the
transformed
data
failed
the
normality
test
for
maximum
%
hemorrhaging,
the
ANOVA
was
performed
on
the
ranked
data.
The
ANOVA
results
on
both
the
raw
data
and
the
rank
transformed
data
are
summarized
in
Table
4­
31.

Table
4­
31.
ANOVA
Results
From
PROC
MIXED
 
Study
III
Field
Nest
Data
Parameter
Test
of
Parameter
Significance
(
p­
value,
raw
data)
Test
of
Parameter
Significance
(
p­
value,
ranks)

Max
%
edema
Max
%
lordosis
Max
%
hemor.
Max
%
edema
Max
%
lordosis
Max
%
hemor.

Treatment
NA
NA
0.3360
NA
NA
0.3703
Stream(
treatment),
random
NA
NA
**
NA
NA
0.9384
*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
variance
components
estimates
being
zeroes.
NA
means
not
available,
because
all
values
were
zeroes.
39
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEBecause
the
treatment
is
not
significant
for
maximum
%
hemorrhaging,
the
means
test
was
not
performed.

Repeated­
Measures
ANOVA
Due
to
a
large
number
of
missing
values
for
"
age"
in
the
dataset,
the
repeated­
measures
ANOVA
was
not
performed.

4.2.4
Study
III
Egg
Cup
Data
ANOVA
Analysis
The
experimental
design
for
Study
III
Egg
Cup
data,
ignoring
time
effects,
is
summarized
in
Table
4­
32.
Again,
observations
associated
with
the
30
µ
g/
L
selenium
treatment
are
not
included
in
the
analysis.
The
following
multivariate
ANOVA
model
was
performed
on
this
dataset
based
on
the
experimental
design:

R
ijk
=
µ
+
C
i
+
S
j(
i)
+
 ijk
where,

R
ijk
=
%
hatch,
%
survival
(
to
the
third
day),
%
edema
(
maximum
incidence
to
the
third
day),
%
lordosis
(
maximum
incidence
to
the
third
day),
%
hemorrhaging
(
maximum
incidence
to
the
third
day),
%
healthy
(
minimum
incidence
to
the
third
day);

µ
=
overall
mean;

C
i
=
treatment
effect,
i
=
1
to
3
treatments;

S
j(
i)
=
stream
effect,
j
=
1
to
2
streams,
nested
within
treatment,
considered
as
a
random
effect;
and
 ijk
=
random
error.

Again,
tests
for
normality
of
model
residuals
and
homogeneity
of
variance
of
the
response
variables
were
performed
on
this
dataset.
The
results
are
summarized
in
Table
4­
33.

The
missing
p­
values
in
the
above
table
result
from
the
occurrence
of
all
zero
values
for
maximum
%
edema
and
%
lordosis
per
spawn
in
the
dataset.
Although
two
variables
failed
normality
tests
(
max
%
hemorrhaging,
%
healthy)
and
only
one
variable
failed
the
homogeneity
of
variance
tests
(
max
%
hemorrhaging),
the
arc­
sine
square­
root
transformation
was
applied
to
all
variables
in
the
dataset,
and
tests
of
normality
and
homogeneity
of
variance
were
performed
on
the
model
using
the
transformed
data.
The
results
are
summarized
in
Table
4­
34.

Although
only
one
variable
(
maximum
%
hemorrhaging)
failed
the
normality
and
homogeneity
of
variance
tests
in
both
the
raw
data
and
transformed
datasets,
ANOVAs
on
both
the
raw
data
and
the
rank
transformed
data
were
performed.
The
results
are
summarized
in
Table
4­
35.
Because
the
treatment
effects
are
not
significant
for
the
response
variables,
the
means
test
were
not
performed.

Repeated­
Measures
ANOVA
Due
to
predominantly
zero
values
for
the
response
variables
in
this
dataset,
the
repeated­
measures
ANOVA
was
not
performed.
40
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
32.
Experimental
Design
for
Egg
Cup
Data
 
Study
III
Selenium
Treatment
Stream
Spawn
(
nest)
Maximum
Incidence
to
the
Third
Daya
Control
1
1
(
B5)
S1
2
(
B1)
S1
5
1
(
A1)
S1
2
(
B1)
S1
3
(
B1)
S2
Recovering
2.5
µ
g/
L
2
1
(
D2)
S1
2
(
D6)
S1
3
(
A1)
S1
7
1
(
A3)
S1
2
(
A3)
S2
3
(
A4)
S1
4
(
B2)
S1
5
(
A3)
S1
6
(
B3)
S1
7
(
B4)
S1
Recovering
10
µ
g/
L
8
1
(
A1)
S1
2
(
A2)
S1
3
(
A5)
S1
3
No
Data
Recovering
30
µ
g/
L
4
1
(
C1)
S1
2
(
C3)
S1
3
(
C4)
S1
4
(
C6)
S1
5
(
C7)
S1
6
No
Data
a
S1:
multivariate
responses
for
sample
1
(
or
cup
1).

Table
4­
33.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Egg
Cup
Data
Normality
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
%
healthy
p­
value
0.4170
0.4474
**
**
0.0001*
0.0001*

Homogeneity
of
Variance
p­
value
0.6849
0.1543
**
**
<
0.0001*
**

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
are
due
to
all
zero
values
in
the
dataset.
41
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
4­
34.
Tests
of
Normality
and
Homogeneity
of
Variance
 
Study
III
Egg
Cup
Data
(
Arc­
sine
Square­
root
Transformed
Data)

Normality
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
%
healthy
p­
value
0.4781
0.6998
**
**
0.0001*
0.0001*

Homogeneity
of
Variance
p­
value
0.9789
0.1324
**
**
<
0.0001*
**

*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
are
due
to
all
zero
values
in
the
dataset.

Table
4­
35.
ANOVA
Results
From
PROC
MIXED
 
Study
III
Egg
Cup
Data
Factors
%
hatch
%
survival
Max
%
edema
Max
%
lordosis
Max
%
hemor.
%
healthy
p­
value
(
raw
data)

Treatment
0.4628
0.6525
NA
NA
0.2424
NA
Stream(
treat),
random
**
**
NA
NA
**
NA
Spawn(
stream),
random
0.0065*
0.0064*
NA
NA
0.3840
NA
p­
value
(
rank
transformed
data)

Treatment
0.4662
0.5861
NA
NA
0.2424
NA
Stream(
treat),
random
**
**
NA
NA
**
NA
Spawn(
stream),
random
0.0080*
0.0079*
NA
NA
0.0276*
NA
*
Significant
p­
value
at
 
=
0.05.
**
Missing
p­
values
resulted
from
the
negative
variance
components
estimates.
NA
means
not
available
because
all
values
were
zeroes.
42
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTESECTION
5
SUMMARY
AND
CONCLUSIONS
The
statistical
findings
from
Studies
II
and
III
are
summarized
below.
Additional
summaries
of
the
statistical
results
of
selenium
effects
on
progeny
in
Study
II
are
provided
in
Tables
5­
1
and
5­
2,
respectively.

Bluegill
Survival
and
Growth

This
study
found
that
selenium
concentrations
of
2.5
and
10
µ
g/
L
did
not
significantly
influence
survival
or
growth
of
juvenile
and
adult
bluegills.

Adult
Spawning
Activity
and
Effects
on
Progeny
(
Field
Nest
Data)
Study
II
Results

Effects
of
2.5
µ
g/
L
Selenium.
ANOVA
results
on
maximum
%
abnormalities
from
the
Field
Nest
Study
II
indicate
that
2.5
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
larvae
for
one
of
the
three
measures
(
maximum
%
hemorrhaging)
based
on
ranked­
transformed
data
(
p<
0.05,
Table
5­
1).
Repeated
measures
analysis
indicates
a
significant
interaction
between
age
and
treatment
effects
for
%
lordosis.
Subsequent
means
testing
indicates
that
2.5
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
%
lordosis
for
two­
day
old
larvae
using
rankedtransformed
data.


Effects
of
10
µ
g/
L
Selenium.
ANOVA
results
on
maximum
%
abnormalities
from
the
Field
Nest
Study
II
indicate
that
10
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
larvae
for
two
of
the
three
measures
(
maximum
%
edema
 
raw
and
ranked
data;
maximum
%
hemorrhaging
 
raw
data,
Table
5­
1).
Repeated
measures
analysis
and
subsequent
means
testing
indicate
that
10
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
%
edema
(
raw
data)
and
%
hemorrhaging
(
ranked
data).
Repeated
measures
analysis
indicates
a
significant
interaction
between
age
and
treatment
effects
for
%
lordosis.
Subsequent
means
testing
indicates
that
10
µ
g/
L
resulted
in
significant
effects
on
%
lordosis
for
three­
day
old
larvae
using
raw
data.
No
statistically
significant
effects
were
observed
on
%
lordosis
at
10
µ
g/
L
for
two­
day
old
larvae
(
ranked
data)
despite
such
effects
occurring
at
2.5
µ
g/
L
(
see
above).

Study
III
Results

No
statistically
significant
effect
of
selenium
was
found
on
any
of
the
three
abnormality
measures
on
larvae
in
streams
recovering
from
prior
selenium
additions
in
Study
II.
43
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTEAdult
Spawning
Activity
and
Effects
on
Progeny
(
Egg
Cup
Data)
Study
II
Results

Effects
of
2.5
µ
g/
L
Selenium.
ANOVA
results
on
maximum
%
abnormalities
from
the
Egg
Cup
Study
II
indicate
that
2.5
µ
g/
L
selenium
resulted
in
no
statistically
significant
effects
on
larvae
for
any
of
the
four
abnormality
measures
(
Table
5­
2).
Repeated
measures
analysis
indicates
a
significant
interaction
between
age
and
treatment
effects
for
three
of
the
four
measures
(%
edema
 
raw
data;
%
hemorrhaging
 
raw
and
ranked
data;
%
healthy
 
raw
and
ranked
data).
Subsequent
means
testing
indicates
that
2.5
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
%
hemorrhaging
and
%
healthy
for
one­
day
old
larvae
using
the
raw
data.
Analysis
of
the
full
and
partial
time­
series
data
sets
showed
consistent
results
for
age­
dependent
effects
at
2.5
µ
g/
L.


Effects
of
10
µ
g/
L
Selenium.
ANOVA
results
on
maximum
%
abnormalities
from
the
Egg
Cup
Study
II
indicate
that
10
µ
g/
L
selenium
resulted
in
statistically
significant
effects
on
larvae
for
two
of
the
four
abnormality
measures
(
maximum
%
edema
 
raw
and
ranked
data;
maximum
%
healthy
 
raw
and
ranked
data,
Table
5­
2).
Repeated
measures
analysis
and
subsequent
means
testing
using
the
full
time­
series
data
indicate
that
10
µ
g/
L
selenium
had
a
statistically
significant
effect
on
%
edema
(
two­
through
five­
day
old
larvae
 
raw
data),
%
hemorrhaging
(
one­
through
three­
day
old
larvae
 
raw
data;
one­
through
four­
day
old
larvae
 
ranked
data),
and
%
healthy
(
one­
through
five­
day
old
larvae
 
raw
and
ranked
data).
For
abnormality
measures
that
did
not
show
a
significant
age
and
treatment
interaction,
repeated
measures
analysis
indicates
that
statistically
significant
effects
occurred
at
10
µ
g/
L
for
%
edema
(
ranked
data)
using
both
the
full
and
partial
time­
series
data
and
%
lordosis
(
ranked
data)
using
the
full
time
series
data.
For
days
common
to
the
two
time
series,
analysis
of
the
full
and
partial
time­
series
data
sets
generally
showed
results
with
respect
to
age­
dependent
effects
at
10
µ
g/
L.


No
statistically
significant
effect
of
selenium
was
found
on
%
hatch
or
%
survival
of
larvae
in
egg
cups
in
Study
II.

Study
III
Results

No
statistically
significant
effect
of
selenium
was
found
on
any
of
the
four
%
abnormality
measures
on
larvae
in
streams
recovering
from
prior
selenium
additions
in
Study
III.
Similarly,
no
statistically
significant
effect
of
selenium
was
found
on
%
hatch
or
%
survival
of
larvae
in
the
recovering
streams
of
Study
III.
44
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
5­
1.
Summary
of
Statistical
Results
from
Study
II
Field
Nest
Dataa
Statistical
Method
(
Factor)
%
edema
%
lordosis
%
hemorrhaging
Raw
Data
Ranked
Data
Raw
Data
Ranked
Data
Raw
Data
Ranked
Data
ANOVAb
(
Treatment)
S,
10
µ
g/
L
(
p
=
0.020)
S,
10
µ
g/
L
(
p
=
0.020)
NS
NS
NS
(
p
=
0.057)
S,
2.5
µ
g/
L
(
p
=
0.022)
S,
10
µ
g/
L
(
p
=
0.014)

Repeated­
Measures
(
Treatment)
S,
10
µ
g/
L
(
p
=
0.027)
NS
­
c
­
c
NS
S,
10
µ
g/
L
(
p
=
0.041)

Repeated­
Measures
(
Treatment
×
Age)
NS
NS
S,
10
µ
g/
L
age
3
(
p
=
0.035)
d
S,
2.5
µ
g/
L
age
2
(
p
=
0.031)
e
NS
NS
a
S
=
statistically
significant
(
p

0.05).
When
statistically
significant
effects
are
reported,
p­
values
are
from
Dunnett's
means
test
unless
otherwise
noted.
NS
=
not
statistically
significant
(
p>
0.05.
p­
values
were
reported
if
p

0.10).
Unless
otherwise
noted,
Tukey's
and
Dunnett's
showed
consistent
overall
indication
of
statistical
significance.
b
ANOVA
was
performed
on
maximum
%
abnormality
per
nest.
c
Statistically
significant
p­
value
was
found
at
 
=
0.05
for
age
×
treatment
interactions.
d
This
p­
value
is
from
Tukey's
means
test,
Dunnett's
test
results
are
not
reported.
e
Tukey's
test
of
the
means
was
marginally
not
significant
(
p
=
0.069).
45
­
DRAFT
FOR
PEER
REVIEW­
­
DO
NOT
CITE
OR
QUOTETable
5­
2.
Summary
of
Statistical
Results
from
Study
II
Egg
Cup
Dataa
Statistical
Method
(
Factor)
%
edema
%
lordosis
%
hemorrhaging
%
healthy
Raw
Data
Ranked
Data
Raw
Data
Ranked
Data
Raw
Data
Ranked
Data
Raw
Data
Ranked
Data
ANOVAb
(
Treatment)
S
(
10)

(
p
=
0.001)
S
(
10)

(
p
=
0.030)
NS*

(
p
=
0.060)
NS*

(
p
=
0.055)
NS*

(
p
=
0.084)
NS*

(
p
=
0.065)
S
(
10)

(
p
=
0.010)
S
(
10)

(
p
=
0.040)

Repeated­
Measuresc
(
Full,
Treatment)
­
f
S
(
10)

(
p
=
0.032)
NS
S
(
10)

(
p
=
0.032)
­
f
­
f
­
f
­
f
Repeated­
Measuresc
(
Full,
Treat.
×
Day)
S
(
10,
Day
2)

(
p
=
0.008)

S
(
10,
Days
3,

4,
5)

(
p
=
0.000e)
NS*

(
p
=
0.080)
NS
NS*

(
p
=
0.086)
S
(
2.5,
Day
1)

(
p
=
0.005)

S
(
10,
Day
1)

(
p
=
0.001)

S
(
10,
Day
2)

(
p
=
0.0004e)

S
(
10,
Day
3)

(
p
=
0.000e)
S
(
10,
Day
1)

(
p
=
0.010)

S
(
10,
Days
2
&
3)

(
p
=
0.000e)

S
(
10,
Day
4)

(
p
=
0.030e)
S
(
2.5,
Day
1)

(
p
=
0.004)

S
(
10,
Day
1)

(
p
=
0.000)

S
(
10,
Days
2,
3,

4,
5)

(
p
=
0.000e)
S
(
10,
Day
1)

(
p
=
0.000)

S
(
10,
Days
2,
3,

4)

(
p
=
0.000e)

S
(
10,
Day
5)

(
p
=
0.006e)

Repeated­
Measuresd
(
Partial,
Treatment)
­
f
S
(
10)

(
p
=
0.017)
NS*

(
p
=
0.062)
NS
­
f
­
f
­
f
­
f
Repeated­
Measuresd
(
Partial,
Treat.
×
Day)
S
(
10,
Day
2)

(
p
=
0.024)

S
(
10,
Day
3)

(
p
=
0.0001e)
NS
NS
NS
S
(
2.5,
Day
1)

(
p
=
0.016)

S
(
10,
Day
1)

(
p
=
0.008)

S
(
10,
Day
2)

(
p
=
0.004e)

S
(
10,
Day
3)

(
p
=
0.0001e)
S
(
10,
Day
1)

(
p
=
0.0122)

S
(
10,
Days
2,
3)

(
p
=
0.000e)
S
(
2.5,
Day
1)

(
p
=
0.039)

S
(
10,
Day
1)

(
p
=
0.009)

S
(
10,
Days
2,
3)

(
p
=
0.000e)
S
(
10,
Day
1)

(
p
=
0.0002)

S
(
10,
Days
2,
3)

(
p
=
0.000e)

a
S
=
statistically
significant
(
p

0.05).
p­
values
are
from
Dunnett's
means
test
unless
otherwise
noted.

NS
=
not
statistically
significant
(
p>
0.10).

NS*
indicates
0.05<
p

0.10.

Numbers
in
parentheses
(
e.
g.,
10)
indicate
treatments
where
statistically
significant
difference
was
observed
relative
to
control.

Unless
otherwise
noted,
Tukey's
and
Dunnett's
showed
consistent
overall
indication
of
statistical
significance.

b
ANOVA
was
performed
on
maximum
%
abnormality
and
minimum
%
healthy
per
cup.

c
Repeated­
measures
analysis
on
"
full"
time­
series
(
Day
1
data
were
excluded
for
%
edema
and
%
lordosis).

d
Repeated­
measures
analysis
on
"
partial"
time­
series
(
Days
1,
4,
and
5
data
were
excluded
for
%
edema
and
%
lordosis;
Days
4
and
5
were
excluded
for
%

hemorrhaging
and
%
healthy).

e
This
p­
value
is
from
Tukey's
test,
Dunnett's
tests
were
not
reported.
All
the
ages
indicated
in
parentheses
have
this
same
p­
value.

f
Statistically
significant
p­
value
was
found
at

=
0.05
for
treatment
×
day
interactions.
