Understanding academic procrastination: A Longitudinal analysis of procrastination and emotions in undergraduate and graduate students

“You cannot escape the responsibility of tomorrow by evading it today”.

Abraham Lincoln.

Abstract

The research presented in this paper examined the relationships between academic procrastination and learning-specific emotions, and how these variables predict one another over time among undergraduate (n = 354) and graduate students (n = 816). Beyond findings showing expected valences of relations between procrastination and positive emotions (enjoyment, hope, and pride) and negative emotions (anger, anxiety, shame, hopelessness, boredom, and guilt), autoregressive cross-lagged panel analyses showed various directional relations between procrastination and emotions over time. More precisely, specific emotions were found to influence procrastination (e.g., undergraduates: anxiety; graduate students: hope), procrastination was found to influence specific emotions (e.g., undergraduates: guilt; graduate students: boredom), and bidirectional relations between procrastination and learning-related emotions were also observed (e.g., graduate students: enjoyment, anxiety, and guilt). Implications for future research on academic procrastination and remedial procrastination interventions for students are discussed.

Appendix

Parceling is frequently used in multivariate analyses involving a latent-variable approach where several items (i.e., indicators) are added together to measure a theoretical construct (Little et al., 2002). By aggregating items together, parceling allows for fewer indicators (reducing the error), and has other benefits including more information in the resulting model (i.e., model efficiency), higher reliability, higher communality, more true-score variance, a higher ratio between the common-to-unique factor variance, as well as optimized sample size to parameter ratios, and better goodness of fit indices (Little, 2013; Matsunaga, 2008). Alongside the benefits associated with parceling, two main disadvantages are consistently cited (Marsh et al., 2013). As noted by Matsunaga (2008), study findings are mixed as to whether or not parceling increases estimation bias in simulation studies by way of decreasing effect size estimates. Well-conditioned data (e.g., normal data with no correlated errors) does not appear to benefit from the use of parceling due to a lack of space for improvements, whereas studies that do not include well-conditioned data have been found to benefit from the reduced error (Matsunaga, 2008). Critics further note that the dimensionality of a scale must be understood if one opts to use parcels, with authors suggesting that parceling may be acceptable when scale items are unidimensional in nature (Little et al., 2013) as the dimensionality of the factors may become distorted (leading to misrepresentations) when parcels are used with multidimensional scales due to potential masking multiple measurement issues (i.e., cross-loading factors, or correlated errors) that are present at the item level. Given that the present data was not perfectly normally distributed, effect size estimates may be marginally inflated from the use of parcels. Moreover, as the dimensionality of each scale was further assessed using EFAs showing all variables to be unidimensional in nature, the possibility of hidden measurement issues when creating parcels was considered minimal. Taken together, parceling was deemed an appropriate method for item reduction in the present study.

Bandalos (2002) found that all-item-parceling (similar to a total score) and three-parcel models showed better goodness-of-fit when compared to six-parcel models. The fewer the parcels, the lower the proportion of error represented, therefore the greater the true variance and model fit. Moreover, it is recommended to use averages of items instead of total scores to ensure that differences in the number of items used in each parcel does not affect the results, making the parcels more comparable (Little, 2013). Thus, the present study utilized parceling as a method of aggregating items within the unidimensional procrastination and emotion scales reducing the number of parameters required to be estimated in each cross-lagged model. The three-parcel method utilizing the random approach was adopted for all main analyses as it represents the most efficient and parsimonious parceling method (See Tables 6 and 7).

Table 6 Longitudinal measurement invariance (undergraduate students)

Table 7 Longitudinal measurement invariance (graduate students)