10
One advantage of using standardized regression coefficients is to provide a common scale
to compare the prediction power of predictors in the same regression equation. For
example, PSAT Math has more prediction power of SAT Math compared with PARCC
ALG02 based on the sample 5 results; PARCC ELA10 and PSAT verbal have similar
prediction power of SAT Reading based on sample 8 results; PARCC ELA10 has more
prediction power of SAT Writing than PSAT Verbal and Writing as shown in sample 10.
Concordance Relationships between Test Scores on PARCC and SAT/ACT
To demonstrate the linkage of scores on different tests, an equipercentile linking
procedure was carried out for pairwise matched samples between PARCC ALG01 and
SAT Math, PARCC ALG02 and SAT Math, PARCC ELA10 and SAT Reading, PARCC
ELA10 and SAT Writing, and PARCC ALG02 and ACT Math. The common-group
design was used in linking based on the 12th graders’ first attempt scores on the tests.
Due to the extremely small sample sizes for the matched groups of students taking both
tests as shown in Table 5, such concordance relationship could not be established for
PARCC ELA10 and ACT Reading, PARCC ELA10 and ACT English, and PARCC
ALG01 and ACT Math.
Equipercentile linking based on the matched samples was carried out using the
software program, Linking with Equivalent Group or the Single Group Design,
abbreviated as LEGS (Kolen & Brennan, 2004). The reported scale scores were used to
link the pairwise tests listed above. After specifying the format of the data input,
subgroup information, input data file names, smoothing values, the score range for the
old test form and the truncation choice, the program conducts equipercentile linking and
outputs the results in the window. In Appendix A, a screenshot captures the input window
for linking the PARCC ALG02 and SAT Math tests. Two smoothing values were
compared in post-linking: 0.3 and 1. The choice of using smoothing values is supported
by the results from simulation studies that the smoothed results outperform non-smoothed
method in reducing linking errors when the population test scores are in fact smooth (Cui
& Kolen, 2009; Hanson et al., 1994). The results with a smoothing value of 1 are
presented in this report due to the fact that after rounding there was little difference
between the results based on the two smoothing values. As some scale scores were not
present in the matched samples, extrapolation had to be done. A scatterplot was generated
to examine the relationship between each pair of tests linked. Based on the total variance
explained and visual inspections of the scatterplots of scores between two linked tests, a
prediction equation was developed using Excel’s best fitting line function. Using the
prediction equation, values not presented in the matched samples were extrapolated. The
extrapolation equations for all pairs of linked tests are presented in Table 10. One thing
worth of note is that when doing extrapolation for PARCC ALG02 and ACT Math based
on the prediction equation, linked ACT scores for PARCC ALG02 scores 784-799 are
actually lower than the linked ACT score for a PARRC ALG02 score of 783 yielded
from LEGS. As ACT Math scores should keep ascending as PARCC ALG02 scores