Science Education News (SEN) Journal 2018 Science Education News Volume 67 Number 1 | Page 28

ARTICLES

and longitudinal differences at the 95 % confidence interval using ESCI software ( Cummings , 2011 ). Each cohort was then examined for data normality , with box plot outliers being deleted from the cohort .

The analytical methods employed were used to examine the contribution of the trial science exam ( independent variable ) to the SC science exam ( dependent variable ) in cross sectional and longitudinal comparisons . This relationship is a proxy measure of the influence that the different lesson plans had upon the memory formed of science during classroom learning ( the trial science exam ) and its subsequent recall and application during the SC science exam . Bivariate and stepwise multivariate regressions using SPSS 17 were used to examine these relationships ( Pallant , 2010 ), with the multivariate regressions being used to explore the contribution of additional contributors ( reading , writing , numeracy and literacy scores ) and isolate the contribution of each significant contributor to the SC science exam in the Postconventional and Postmeta cohorts . All regressions had settings of two-tailed a = . 05 , the prediction interval for the mean was set at 95 %, missing multivariate data was excluded listwise and the stepping probability criteria for data entry was F = . 05 and F = . 10 for data removal . Regression and residual casewise diagnostics for outliers was set at 2 standard deviations . The data coherency of all regressions was validated using split regressions ( Snee , 1977 ) using the SPSS random number generator with a fixed value of 552804 . The f 2 effect sizes of multivariate contributors were determined using ClinTools ( Devilly , 2007 ) and then converted to their Cohen ’ s d effect size equivalent . Differences in the correlation coefficients and the d effect sizes were examined at the 95 % confidence interval using ESCI software , with the achieved power of the difference in bivariate slopes determined using G * Power ( v . 3.1.3 ) ( Faul , Erdfelder , Buchner , & Lang , 2009 ).

generate normality .

Figure 1 Difference in means of trial and SC science exam scores .

1 2 3 4 5 6 7 8

Figure2 Difference in means of multivariate scores .

RESULTS

Descriptive data

Generally , the mean exam scores for all cohorts did not differ at the 95 % confidence interval , indicating that science classes with comparable student characteristics were identified for inclusion into each cohort ( see figures 1 and 2 ). The single difference detected was in the mean SC science exam score for the Preconventional vs . Postmeta comparison , suggesting that learning characteristics of these two cohorts were significantly different . However , as their mean trial science exam score was comparable , as were the reading , writing , numeracy and literacy scores of the Postconventional vs . Postmeta comparison , this one difference in descriptive data is minor given the overall trend of the descriptive data being equal . This equality of means was broadly extended to the normality of the data for regression analysis and the residuals of the regressions . The single exception was the removal of three data sets from the Postconventional cohort to

28 SCIENCE EDUCATIONAL NEWS VOL 67 NO 1