A table of means, standard deviations and zero-order correlations between all of the study variables for the pooled sample dataset is presented in Table 1. Table 2 presents the same information for the two cohort datasets separately, along with t tests comparing means of the two cohorts. The relationships between key variables in the pre-lecture capture year are presented in the ‘above-diagonal’ of Table 2 and relationships in the post-lecture capture year are presented in the ‘sub-diagonal’.
Briefly summarising bivariate relationships linked to the study’s hypotheses, within the pooled data, we find a significant positive relationship between attendance and final grade on the module (r = 0.416, p < 0.001); this significant positive relationship was found between attendance and all four aspects of the module’s assessment. This association between attendance and final grade is observed in both years, both before lecture capture (r = 0.340, p < 0.001) and after its introduction (r = 0.469, p < 0.001). Thus, students who attended more than others tend to score more highly across the module’s assessments, supporting hypothesis 3a.
We compared the means of the two cohorts across the attendance and attainment measures (see Table 2). The pre-lecture capture cohort showed significantly greater average attendance in the matched 3 weeks of lectures (pre- versus post-lecture capture cohort means of 1.58 versus 1.19; t(319) = 3.12, p < 0.01) and significantly higher coursework grades (pre- versus post-lecture cohort means of 62.82 versus 58.28; t(319) = 3.56, p < 0.001). There were no significant differences between pre- versus post-lecture cohort means for year 1 grade, quiz grade, participation grade, exam grade or final grade.
In the most part, lecture capture usage did not show significant relationships with other measures collected; however, within-term lecture capture viewing was positively related to coursework grade (r = 0.185, p < 0.05) and weeks 4 to 11 lecture attendance (r = 0.178, p < 0.05). Thus, there was a tendency for the people who attend lectures to also view lecture captures during the term; these people also tended to do slightly better at the in-term coursework. The correlations between module-end lecture capture use with the 3-week attendance (r = 0.128, p > 0.05) and the 8-week lecture attendance (r = 0.155, p > 0.05) measures as well as final grade (r = 0.096, p > 0.05) were all non-significant.
In terms of significant correlations involving the control variables, a significant positive relationship was found between year 1 grade and attendance (r = 0.262, p < 0.001). There were also significant positive relationships between year 1 grade and all attainment indicators over the 2 years (r = between 0.362 and 0.562, p < 001). These results show that students who did well in the previous year tended to attend more and achieve greater attainment than those who performed less well in their first year of study. In the pooled data, females score higher on weekly quizzes than males across the 2 years. The results in Table 2 show that gender is positively related to other elements of assessment grade in the first year; females do better at the quiz and participation and get higher final grades before lecture capture is introduced; however, these relationships fall from significance after lecture capture is introduced.
Predicting attendance (dual-cohort analysis Table 3)
A hierarchical regression model predicting lecture attendance (the matched 3 weeks), using average year 1 grade and gender as independent variables, was significant f(2,318) = 13.83, p < 0.001, and accounted for 8% of the variance in attendance (R-square = 0.080). Adding the lecture capture availability year dummy improved the model significantly, f(3,317) = 14.66, p < 0.001, which now accounted for 11.4% of the variance in attendance (R-square = 0.114). Year 1 grade was positively related to attendance (beta = 0.280, p < 0.001) and in the pooled sample there was a gender effect (beta = 0.115, p < 0.05, males attended less than females). The lecture capture availability dummy accounted for a significant portion of variance and resulted in a significant negative beta (beta = − 0.206, p < 0.001), with attendance being lower after the introduction of lecture capture when accounting for gender and general academic ability. Thus, hypothesis 1 is supported.
To get a better understanding of the nature of the relationship between lecture capture availability and attendance, Fig. 1 presents the attendance patterns for the students across the 2 years before and after lecture capture availability. The proportion of the cohort who did not attend any of the three matched lectures rose from 19.9 to 39.4%, about the same proportion attended one lecture (26.1% before lecture capture introduction and 25% after introduction), fewer students attended two lectures (30.4% before lecture capture and 13.1% after) and about the same attended all three (23.6% before lecture capture and 22.5% after).
Predicting attainment (dual-cohort analysis Table 3)
The regression model predicting final grade attainment using average year 1 grade and gender as independent variables was significant, f(2,318) = 109.64, p < 0.001, and accounted for 41% of the variance in student attainment (R-square = 0.408). Adding the lecture capture availability year dummy significantly improved the model f(3,317) = 79.00, p < 0.001, which now accounted for 43% of the variance in student attainment (R-square = 0.428). The lecture capture availability year dummy was significant and negative (beta = − 0.141, p < 0.01) suggesting that lecture capture introduction has a negative effect on overall attainment, over and above the effects of general academic ability. This supports hypothesis 3b.
Adding lecture attendance as an independent variable again significantly improved the model f(4,316) = 73.13, p < 0.001, which now accounted for 48% of the variance in student attainment (R-square = 0.481). Year 1 grade was strongly (positively) related to final grade on the module (beta = 0.582, p < 0.001), with no gender effect (beta = 0.024, NS); lecture attendance showed a significant positive relationship with attainment (beta = 0.245, p < 0.001); the addition of this measure accounted for an extra 6.5% of the variance in student grades. Thus, hypothesis 3a is supported. The lecture capture availability year remains significant and negative (beta = − 0.091, p < 0.05), suggesting that lecture capture introduction continues to have a negative effect on overall attainment over and above the impact of attendance.
The same analyses were conducted with exam grade as a dependent variable and the findings are largely the same as with final grade (with one small exception). Attendance showed a significant positive relationship with attainment (beta = 0.161, p < 0.001) and the addition of this measure accounted for an additional 3.2% of the variance in student exam grades. Although significant before adding attendance (beta = − 0.101, p < 0.05), the lecture capture availability year dummy was no longer significant once accounting for lecture attendance (beta = − 0.068, p > 0.05).
We assessed the mediation effect of lecture capture availability on attainment through attendance using the Process Macro (Hayes 2008). This calculates an indirect effect coefficient to represent the mediation using bias-corrected bootstrapped sampling. The analysis revealed a significant indirect effect through attendance in the relationship between lecture capture availability and final grade (beta = − 0.051, p < 0.05, LLCI = − 0.089:ULCI = − 0.022), which supports the proposed mediation and hypothesis 3c. A similar significant indirect effect is also found for exam grade (beta = − 0.033, p < 0.05, LLCI = − 0.069:ULCI = − 0.013), although as noted above within this model the effect of lecture capture availability becomes non-significant when attendance is accounted for, which suggests a more complete mediation. The resulting model is represented in Fig. 2.
Analyses incorporating lecture capture viewings
We tested the study’s hypotheses that included an aspect of lecture capture usage with zero-order correlations and partial correlations for hypothesis 2 and regression analyses for hypotheses 4a/4b and 5 with the 2016 cohort. As mentioned above, the zero-order correlation between lecture capture viewings and lecture attendance post-lecture capture introduction was positive but non-significant (r = 0.128, NS); the partial correlation between lecture capture viewing and lecture attendance after accounting for our gender and year 1 grade (our controls) was also non-significant (r = 0.115, NS). Thus, hypotheses 2 was not supported. Two regression models were tested predicting both exam and final grades using average year 1 grade, gender, attendance of lectures over 3 weeks and total lecture capture views (including revision period) as independent variables. The full model predicting exam grade was significant (F(4,154) = 23.97, p < 0.001) and accounted for 38.2% of the variance in exam attainment (R-square = 0.382), see Table 4. The model predicting final grade was also significant (F(4,154) = 45.62, p < 0.001) and accounted for 52.9% of the variance in 2016/2017 student attainment (R-square = 0.586), see Table 4.
Adding lecture capture use showed no significant increase on the variance accounted for in attainment; thus, the spirit of hypothesis 4b is supported in that lecture capture use is not a predictor of attainment when controlling for ability. However, as the positive zero-order correlation between module-end lecture capture use and grade does not reach significance, the first part of this hypothesis is not supported (hypothesis 4a). Given the significant correlation between lecture capture usage and coursework grade, we ran a further regression using coursework grade as a dependent variable and found that after controlling for general academic ability, gender and attendance, lecture capture usage did not significantly predict coursework grade (beta = 0.051; p > 0.05). Therefore, we find no evidence that lecture capture usage has any unique effect on student attainment.
To test hypothesis 5, we ran a moderation test using Process (Hayes 2008); the two tests involved adding an interaction term (the multiplication of attendance and lecture capture usage) to the third and sixth model in Table 4. In both cases, the interaction term was non-significant (interaction-term beta = − 0.003, p > 0.05 predicting exam grade; beta = − 0.030, p > 0.05 predicting final grade). Thus hypotheses 5 is not supported.
Student behaviour profiles and attainment 2016/2017
To examine the lack of interaction between lecture capture usage and attendance in more detail, we grouped the students into three profiles of weeks 4–11 lecture attendance behaviours: a group that never attended lectures (30%), a group that attended between one and four lectures (41.9%), and one that attended more than 50% of lectures (28.1%). We combined our three categories of attendance behaviour with the three profiles of lecture capture views to explore the potential patterns of attainment across these different profiles. As Fig. 3 shows, the mean grade for the low-attendance/high lecture capture use group (mean = 63.77) is slightly higher than the low attendance/low or mid lecture capture use groups. Further examination showed that this group only had seven students in, three of which had mean final grades above 65 (71.4, 68.9, 67.8). The fact that only three students in the group showed high grades but very low attendance and high lecture capture use indicates that if lecture capture can help some students recover from non-attendance, these only represent a very small proportion of the cohort (under 2%). This perhaps explains why the moderation tests are not significant (despite the appearance of an interaction with Fig. 3).