The link - e

advertisement
e-asTTle linking study
Introduction
The purpose of the linking study was to provide a link between the current and previous
versions of the e-asTTle writing scale. It presents a way of converting e-asTTle writing scores
measured on the previous version of e-asTTle to those measured on the current version.
This document:
ο‚·
outlines how this conversion was determined
ο‚·
provides a brief description of the utility of the marking rubric for the current
version of e-asTTle writing.
The link
This section describes the link between the previous and current e-asTTle writing scales.
Converting previous e-asTTle scores to current e-asTTle 𝒂𝑾𝒔 scores
In order to convert a previous e-asTTle scale score above 1250 to a current e-asTTle scale score,
apply the following formula:
π‘Žπ‘Šπ‘  ≈ 0.95 ∗ π‘Š + 95
Where:
ο‚Ÿ
π‘Žπ‘Šπ‘  denotes the current e-asTTle writing score
ο‚Ÿ
π‘Š denotes the previous e-asTTle writing score
This is summarised in tabular form in Table 1.
Scores will be less precise after they have been converted. An estimate of error has been
calculated by combining the standard error associated with linking with the measurement error
in the previous and current versions of e-asTTle. The standard error on a score converted from
the previous version of e-asTTle is around 70 aWs points.
A previous e-asTTle score of 1250 or lower should not be converted.
Page 1 of 7
Converting current e-asTTle πšπ–π¬ scores to previous e-asTTle scores
In order to convert a current e-asTTle scale score above 1280 to a previous e-asTTle scale score,
apply the following formula:
π‘Š ≈ 1.05 ∗ π‘Žπ‘Šπ‘  + 100
Where:
ο‚Ÿ
π‘Žπ‘Šπ‘  denotes the current e-asTTle writing score
ο‚Ÿ
π‘Š denotes the previous e-asTTle writing score
Scores will be less precise after they have been converted. An estimate of error has been
calculated by combining the standard error associated with linking with the measurement error
in the previous and current versions of e-asTTle. The standard error on a score converted from
the current version of e-asTTle is around 73 previous e-asTTle score points.
A current e-asTTle score of 1280 or lower should not be converted.
Table 1 Table of scale conversions
Previous e-asTTle scale score (π‘Š)
Previous e-asTTle scale score (π’‚π‘Šπ‘ )
1250
1283±70
1275
1306±70
1300
1330±70
1325
1354±70
1350
1378±70
1375
1401±70
1400
1425±70
1425
1449±70
1450
1473±70
1475
1496±70
1500
1520±70
1525
1544±70
1550
1568±70
1575
1591±70
1600
1615±70
1625
1639±70
1650
1663±70
1675
1686±70
1700
1710±70
1725
1734±70
1750
1758±70
1775
1781±70
1800
1805±70
Page 2 of 7
Methodology
The linking was determined by the data obtained from two different approaches to creating a
link: equipercentile equating and linking determined by double marking previous e-asTTle
scripts. These two approaches were implemented separately and are described separately
below. The data they generated was pooled and then subjected to a linear regression analysis.
This is also described separately.
Equipercentile linking
The data enabling seven equipercentile links between the previous and current versions of easTTle writing was created – one link for each of the Year 4, Year 5, Year 6, Year 7, Year 8, Year 9
and Year 10 student populations. This data was drawn from the normative information provided
by the current and previous e-asTTle tools. Specifically, the points below which 0.1%, 10%, 20%,
30%, 40%, 50%, 60%, 70%, 80%, 90% and 99.9% of the student populations for a given year-level
scored on each version of e-asTTle writing were generated. These are depicted in Figure 1.
Figure 1 The equipercentile link data
2000
1900
1800
Current e-asTTle score
1700
1600
1500
1400
1300
1200
1100
1000
1000
Year 4
1200
Year 5
1400
Year 6
1600
1800
Previous e-asTTle score
Year 7
Year 8
Year 9
2000
Year 10
Page 3 of 7
This graphic suggests a strong relationship – the data is obviously linear and sits very close to a
line of best fit. This linear behaviour is not surprising given that both scales are transformations
of logit-based writing measurement scales. For example, at Year 6 both scales are constructed so
that the mean score is 1500 and the standard deviation is 100 score points.
Linking based on double-marking
Sixty-nine student scripts (responses to e-asTTle prompts) – already marked using the previous
e-asTTle writing rubric – were re-marked using the current e-asTTle rubric. This was
implemented independently by two of the developers of the current e-asTTle writing rubric.
These two developers were asked to justify or re-mark any outliers.
Three student scripts with previous e-asTTle scores below around 1250 were removed from the
data. This was because the relationship between the previous and current e-asTTle scores for
these scripts was markedly different from that for the remaining scripts. The reason for this
difference may be due to the current e-asTTle rubric being able to distinguish responses at lower
levels of writing proficiency than the previous e-asTTle rubric. The resulting data is depicted in
Figure 2. For comparative purposes, the data including the removed data points is depicted in
Figure 3.
Figure 2 The student script data
2000
1900
1800
Current e-asTTle score
1700
1600
1500
1400
1300
1200
1100
1000
1000
1100
1200
1300
1400
1500
Previous e-asTTLe score
Marker 1
1600
1700
1800
Marker 2
Page 4 of 7
Again, the linear relationship in this data seems fairly clear. Some of the data sits further from a
line of best fit than that depicted in Figure 1 but this is to be expected given that variance
between markers has been introduced.
Figure 3 The student script data including the removed points
2000
1900
1800
Current e-asTTle score
1700
1600
1500
1400
1300
1200
1100
1000
1000
1100
1200
1300
1400
1500
1600
1700
1800
Previous e-asTTLe score
Marker 1
Marker 2
It is clear that the data points in Figure 3 which have previous e-asTTle scores less than around
1250 are demonstrating a relationship different from the remaining data points.
Linking using the pooled data
The equipercentile links and the script data were pooled and a line of best fit was overlaid on the
pooled data. The line of best fit from the pooled data was used to determine the link between
the previous and current e-asTTle writing scales. This is depicted in Figure 4.
Page 5 of 7
Figure 4 The pooled linking data
2000
1900
y = 0.9534x + 94.542
R² = 0.8246
1800
Current e-asTTle score
1700
1600
1500
1400
1300
1200
1100
1000
1000
1100
1200
1300
1400 1500 1600 1700
Previous e-asTTle score
Markers and norms
1800
1900
2000
Linear (Markers and norms)
The pooled data agrees closely with its linear model as is indicated by the high R-squared value.
An estimate of the standard error associated with linking – around 50 π‘Žπ‘Šπ‘  – was estimated by
using the standard error of the intercept parameter in the linear regression performed to find
the link formula.
Utility of the marking rubric
The following observations were made by the two developers of the current e-asTTle writing
rubric who marked the student scripts used in this linking study. They focus on the utility of the
current e-asTTle marking rubric.
Page 6 of 7
ο‚Ÿ
The current marking rubric has been informed by close observation, marking and moderation
of student writing from Year 1 to Year 10. It therefore reflects a broader range of writing
proficiency than the previous e-asTTle writing rubrics. In particular, it can recognise those
students who are just emerging as independent writers, allowing the achievement of all
independent writers to be assessed.
ο‚Ÿ
The current rubric is based on the principle of recognising student achievement in seven
elements of writing. It contains detailed, precise descriptions of skill development in each
element. Because the writing used to develop them was produced by students from Years 1
to 10, the descriptions capture the early stages of development in each element. The rubric
can therefore provide meaningful diagnostic information for any student who is able to
communicate at least one or two simple ideas independently. Information at this level was
not available in the previous version of e-asTTle writing.
ο‚Ÿ
Because the current rubric involves assigning ‘Rubric Scores’ in each element rather than
curriculum levels, there is the possibility of greater accuracy in scoring decisions. Rather than
first placing a student’s writing within a broad curriculum level then considering finer
distinctions within this (ie, the BPA system of the previous version of e-asTTle), the current
rubric allows for a consideration of observable skill levels within each element. This
consideration is independent of the curriculum. It is the student’s overall performance that is
compared with the curriculum.
ο‚Ÿ
The current e-asTTle marking rubric is quite robust. That is, it can be applied effectively to
writing produced in response to previous e-asTTle writing prompts, as long as the purpose for
the writing is to describe, narrate, recount, explain or persuade. Similarly, it can be applied
effectively to any writing prompts comparable to the current or previous e-asTTle writing
prompts. This is because the elements of the current rubric focus on observable indicators of
writing proficiency. The scores from prompts that are not part of e-asTTle should not be
entered into the tool itself.
ο‚Ÿ
Although some of the elements of the current e-asTTle marking rubric do not translate
directly into the language of the previous marking rubric, the link described in this document
is statistically sound. This indicates that the overall writing scores of the previous and current
versions of e-asTTle are readily able to be translated and underscores the robustness of the
current e-asTTle marking rubric.
Page 7 of 7
Download