Li
Ye
,
Razanne
Oueini
,
Austin P.
Dickerson
and
Scott E.
Lewis
*
University of South Florida – Chemistry, Center for the Improvement of Teaching & Research in Undergraduate STEM Education, 4202 E. Fowler Avenue CHE205, Tampa, Florida 33620, USA. E-mail: slewis@usf.edu; Fax: +1-813-974-3203; Tel: +1-813-974-3099
First published on 27th July 2015
This study used a series of text message inquiries sent to General Chemistry students asking: “Have you studied for General Chemistry I in the past 48 hours? If so, how did you study?” This method for collecting data is novel to chemistry education research so the first research goals were to investigate the feasibility of the technique and the evidence for validity of the data collected. The results showed that text messages provide ample data on students' study habits though initial participant recruitment may pose a challenge. This study also explored evidence for validity and found that the percent of students reporting studying peaked with each exam date matching the expected trend (content validity) and participants in the study had only small departures from the population of students at the setting (generalizable validity). Second, students' study habits were characterized using cluster analysis finding three clusters: students that knowingly do not study, students who describe mandatory course components as studying and students who study in addition to the mandatory course components. These student groups were compared on a common exam in the course with the last group out-performing those who knowingly do not study. Finally, student study habits were charted across the semester and show signs of adapting, possibly as a result of course expectations or course content.
There has been little research attention toward measuring study habits in the context of post-secondary chemistry. Richards-Babb and Jackson (2011) investigated gender differences in study habits through a survey given at the end of General Chemistry and reported that male students were more likely to procrastinate. Also related, Li et al. (2013) investigated post-secondary chemistry students' conceptions of learning chemistry and approaches to learning chemistry. Conceptions of learning chemistry were measured by a survey developed based on earlier, more general research interviewing students about learning experiences. This work identified memorizing, testing, calculating/practicing and higher order thinking labeled as transforming as the relevant themes in students' conceptions of learning. Approaches to learning were measured based on the previously described study processes and the continuum from surface to deep learning. The study found that students who were characterized as deep learners conceived of learning in multiple ways including transforming, memorizing and testing, while learners that used surface strategies employ memorizing and testing.
Most studies that investigated students' study habits used a single-admission survey that may be problematic for two reasons. First, as a single measure, it presumes that students' study habits are constant, whereas it is possible that students' study habits adapt to the nature of the content and with familiarity toward assessment expectations. Second, as an in-class survey, it relies upon retrospection on behalf of the student, particularly when it is given at the end of the semester. Past research has called into question the accuracy of retrospective accounts, particularly at lengthier time intervals (Bernard et al., 1984). By exploring study habits at multiple time points, both problems may be minimized as changes over time can be documented and participants would not be asked to reflect upon several months of study habits.
There have also been efforts made to improve students' study habits. Cook et al. (2013) implemented a one-day lecture for General Chemistry students that presented differences in expectations between post-secondary and secondary education as described by Bloom's taxonomy. They also presented metacognitive learning tools including a study cycle. At the conclusion, students made a brief written statement committing to use some of the tools presented. Student attendance to the lecture was voluntary though students received a bonus equivalent to 0.5% of their grade for attending. Attending students were compared to non-attendees on the metric of points earned in the course post-implementation (transformed to follow a normal distribution). The statistical comparison used students' first exam score as a covariate as it preceded the implementation. The results of an ANCOVA showed that students who attended the treatment performed better on the outcome measure than those who did not. These results indicate the potential importance of student study habits to student learning, but the results could also be indicative of a confounding variable such as student motivation. For example, the authors did note that the control group missed more exams than the treatment group and that could be an indication that the groups differed in their motivation to succeed in the class. Incorporating a measure of study habits before and after the intervention would further elucidate the impact of the intervention and help better establish a causal connection such as the intervention impacted student study habits which led to greater student learning.
In summary, considerable research has shown a relationship between study habits and academic performance in post-secondary education but not in the sciences. The following study seeks to address this gap in the research literature by investigating the frequency and types of study habits in General Chemistry. The creation of a detailed measure of student study habits as described below can open two potentially fruitful areas of study. First, measuring study habits can inform efforts to better understand the factors related to academic success (study habits as an independent variable). Second, measuring study habits can aid explorations of instructional efforts to improve students' study habits (study habits as a dependent variable). The following study takes an exploratory approach to examine study habits as an independent variable.
ESM has been used in a variety of contexts, particularly in the field of psychology where it has been used to explore constructs as diverse as morality, mental illness and substance abuse. (Smyth and Stone, 2003; Hofmann et al., 2014) It has also been used with medical applications to investigate disorders, drug abuse and treatment effectiveness. (Hektner et al., 2007) In education it has been used most often at the secondary level to investigate student motivation, satisfaction with the educational environment or the nature of the environment. (Csikszentmihalyi, 2014) To date, we could not locate a study that has used ESM to explore a post-secondary chemistry setting or post-secondary student study habits.
(1a) To what extent is it feasible to measure student study habits using ESM?
(1b) To what extent is there evidence for the validity of the data collected on student study habits?
(2) Which study habits were related to academic performance in the course as measured by a cumulative final exam?
(3) How did student study habits change over the course of the semester?
To aid student studying, past exams were posted approximately two weeks before the actual exam and are referred to as practice tests. The textbook used was Tro's Chemistry: A Molecular Approach (2014) and the content sequence was: quantum numbers, periodic trends, Lewis structures, shapes and polarity, gas laws, thermodynamics, intermolecular forces and properties of solutions. Grades were determined largely by performance on three in-class exams (15% each) and the cumulative final exam (25%). A smaller portion of the grade was attributed to three different effort-based measures at 10% each. First, the class used weekly peer-led problem-solving sessions where students worked in groups on problems designed by the instructors with the aid of peer-leaders (Lewis, 2011). Attendance and participation in these sessions was worth 10% of their grades. Second, students were graded on their performance on eight online homework assignments using Sapling Learning. Third, instructors used clickers to facilitate in-class questions in the large lecture-hall setting.
Student responses to the text messages were combined with data collected in the normal educational setting from either university records or in-class records. This data includes student responses to the revised two-factor Study Process Questionnaire (rSPQ) administered on the first day of class (Biggs, 2001). The rSPQ is a 20 item Likert-scale instrument meant to measure students' study processes. The instrument was revised by the original instrument's author and measures respondents on two sub-scales: deep approach and surface approach. The deep approach can be characterized by intrinsic interest or a motivation to understand. Example items from the rSPQ that measure the deep approach are “I come to most classes with questions in mind that I want answering” and “I find that at time studying gives me a feeling of deep personal satisfaction.” The surface approach can be characterized by a narrow focus on content and memorization with example statements “My aim is to pass the course while doing as little work as possible” and “I find the best way to pass examinations is to try to remember answers to likely questions.” In this study, students were asked to consider their study habits in general, but if they need to consider a subject, consider how they would study for chemistry or a science course. The Likert-scale was a five-point range from “this item is never or only rarely true of me” to “this item is always or almost always true of me,” and each factor score represents the sum of ten associated items.
In this study the rSPQ is thought to measure the quality of studying where the deep approach describes the desirable educational process (Biggs 2001). This is differentiated from study habits, which describe the type and frequency of studying. There are expected relationships between the constructs, for example students who employ a deep approach are expected to study more frequently. Student scores on the rSPQ are considered in contrast to their cohort as recommended by Biggs (2001). Additional measures include student demographics and SAT scores (a measure of incoming college preparation) from university records, and student performance on exams from in-class records. Descriptive statistics on each of these measures for the population of 899 students are presented in Table 1.
Variable | Average | St. Dev. | N | Theoretical range | Cronbach's α |
---|---|---|---|---|---|
SAT math | 552 | 68 | 695 | [200, 800] | |
SAT verbal | 548 | 73 | 695 | [200, 800] | |
Deep approach | 32.1 | 6.7 | 797 | [10, 50] | 0.826 |
Surface approach | 23.8 | 6.2 | 797 | [10, 50] | 0.776 |
Final exam | 48.5 | 15.7 | 754 | [0, 100] | 0.816 |
In terms of feasibility, it is plausible to have a substantial portion of the recruited population respond to this data collection technique. It is worth noting that the raffle incentive required responses to 80% of the inquiries (23 or more inquiries), which may partially explain the rise in number of participants who responded to 23 or more inquiries. Instead, the largest source of data attrition in the study was during the initial recruitment, where out of an initial population of 670 students, 301 students agreed to participate (44.9%). This suggests that future research studies that intend to rely on a large number of responses would benefit by planning for a substantially larger recruitment pool. The current data indicates that roughly one-quarter of the initially recruited population provided responses to at least half of the inquiries.
At the close of the semester, an additional text message inquiry was sent asking participants if they would participate in a similar study using text messages in the future. Of the 94 respondents, 78% responded positively compared to 18% negative (with the remainder unsure). The most common negative comment (5 responses) was that students found the messages annoying. However, the most common comment (52 responses) described the convenience in participating with some indicating it was less of a time commitment compared to traditional studies.
Participants average (St. Dev.) N | Non-participants average (St. Dev.) N | |
---|---|---|
a URM = under-represented minority (as defined by the National Science Foundation). | ||
SAT Math | 554 (65) 237 | 550 (69) 458 |
SAT Verbal | 552 (70) 237 | 545 (74) 458 |
Deep Approach | 32.7 (6.6) 288 | 31.8 (6.7) 509 |
Surface Approach | 23.6 (6.3) 288 | 23.9 (6.2) 509 |
Participant demographics (N) | Non-participant demographics (N) | |
Gender | 65% Female (300) | 59% Female (594) |
Minority | 45% URMa (277) | 40% URM* (559) |
To explore the content of the responses, the text messages were coded using an open-coding scheme. The coding process resulted in 16 codes as shown in Table 3. Each response was coded and responses could be coded with multiple codes. For example, “Yes the back of book problems, reading the chapter, and doing the online homework assignment” was coded for Textbook, Practice Problems and Homework. To check the inter-rater reliability of the coding scheme, 10% of the text messages were randomly selected and coded by a researcher who was independent of the first coding pass. The resulting codes agreed with the original code for 94% of the responses. Table 3 also presents the relative frequency of the codes as the percent of responses that used a particular code.
Code | Percent (%) |
---|---|
Did not study | 42.2 |
Reviewed notes or PowerPoint | 18.8 |
Reviewed the textbook | 16.4 |
Online homework | 14.2 |
Practiced problems | 6.8 |
Previous exams or study guides | 5.7 |
Unspecified yes | 4.5 |
Used online materials | 2.6 |
Worked with friends or in a group | 2.4 |
Attended peer leading or reviewed peer leading assignment | 2.1 |
Worked with a tutor | 1.9 |
Attended class | 1.1 |
Made flashcards | 0.9 |
Visited professor | 0.3 |
Attended lab | 0.2 |
Reviewed tables, models or charts | 0.2 |
Text messages were then also coded dichotomously as either a study habit was used or the participant did not study. The codes unspecified yes, attended class, attended peer leading or attended lab were coded as missing in this categorization as it was not clear whether these participants had employed a study habit. With the new dichotomous codes, the percent of participants employing a study approach was determined for each text message inquiry. The percent of participants using a study habit is plotted by date in Fig. 2, using only those participants who responded to at least half of the 28 inquiries (N = 188). The vertical lines in Fig. 2 correspond to the exam dates in the class. From Fig. 2, the percent of students who report studying increases leading up to each exam, peaks at the exam date and subsequently drops-off. This matches the expected pattern of instructional experience where student inquiries tend to ramp up leading up to the date of an exam, lending content validity to the responses received.
Ultimately, the measure of students' study habits proposed is still reliant on self-report. Self-reported data may be influenced by factors that cannot be ruled out such as participants' belief in a socially desirable response pattern or errors in participants' efforts to recollect. Such factors would impact the accuracy of the responses as a measure of actual student actions. As study habits by definition occur outside of a controlled research setting, attempts to triangulate the measure without relying on self-report would require extensive observations that would impose on participants' privacy. This serves as an unavoidable limitation of this study though it is proposed that participants' self-report of study habits do offer value in understanding the factors needed for successful academic performance.
Initially, correlations between each study habit and the final exam were conducted. Each correlation indicated a weak relationship with the strongest relationship of 0.14 between percent of responses using the textbook and final exam score. Since correlations only indicate the strength of a linear relationship, the data was further explored for the possibility of relationships that do not follow a linear pattern. Owing to the substantial number of study habits present, the decision was made to conduct a cluster analysis to look for patterns among the multiple study habits. Cluster analysis is an algorithm that measures the distance between each case (student) on the variables (frequency of study habits) and combines pairs of students who feature the smallest distance into a cluster. The algorithm continues to combine students and clusters of students until it reaches a user-specified number of clusters. In this way, cluster analysis can be used to find groups of students who have similar profiles across multiple variables (Everitt et al., 2011). Cluster analysis can be used to describe the data in terms of number of students per group and the average study habits within each group. These groupings can then facilitate investigating relationships among other measures.
For the cluster analysis, only the six most prevalent study habits in Table 3 were used, as these were each represented by at least 5% and were also readily interpretable (the next most prevalent code would be the unspecified yes). A hierarchical cluster analysis using Ward's method and squared Euclidean distance was employed to create clusters that were distinct from each other (Aldenderfer and Blashfield, 1984). To determine the number of clusters, the cluster analysis began with six clusters that were evaluated based on sample size in each cluster and the average percent for each study habit. Then an analysis to create five clusters was conducted to determine which two clusters were combined; these clusters were evaluated based on qualitative similarity on study habit percentages and the relative sample size of each cluster. The analysis was continued until reducing the number of clusters meant losing a cluster that was substantially distinct. The intent was to determine the number of clusters that led to a reasonable representation in the sample for each cluster and where each cluster was distinct. This resulted in three clusters that are characterized in Table 4.
Cluster 1 | Cluster 2 | Cluster 3 | |
---|---|---|---|
Bold indicates study habit has more than +0.5 standard deviation different than the overall average; italic text indicates study habit is less than −0.5 standard deviations different than the overall average. | |||
Sample size | 64 | 62 | 62 |
Study habits | Average (St. Dev.) | ||
---|---|---|---|
Did not study | 67% (13%) | 26% (14%) | 33% (15%) |
Reviewed notes or PowerPoint | 8% (9%) | 22% (22%) | 26% (16%) |
Reviewed the textbook | 8% (11%) | 35% (16%) | 8% (8%) |
Online homework | 5% (7%) | 11% (9%) | 25% (16%) |
Practiced problems | 4% (6%) | 10% (12%) | 6% (9%) |
Practice tests or study guides | 3% (4%) | 6% (7%) | 9% (8%) |
Table 4 describes three distinct clusters that arose from the study habits in the sample. To place the values in context, 14% would indicate that they used the study habit at least twice and at most four times over the course of the semester. The sample distribution among the three clusters is relatively even which suggests that each cluster has prevalence among the sample. Participants in Cluster 1 indicated not studying far more often than the rest of the sub-sample (67% versus 42% for the sub-sample) and subsequently indicated reviewing notes and the online homework less often than the sub-sample. Cluster 2 was more than one standard deviation greater than sub-sample on use of the textbook (35% versus 16%). Cluster 2 was also higher on practicing problems and lower on the percentage of not studying. Cluster 3 was noteworthy for describing the online homework as their study habit, but was also higher on reviewing notes and the practice tests.
The three clusters were compared on the five other measures with data presented in Table 5. To compare the clusters an Analysis of Variance (ANOVA) was performed with α = 0.05 which provides a group-wise error rate of 0.23 across the five tests. The effect size was also characterized by Cohen's f, where 0.10 is a small effect and 0.25 is a medium effect (Cohen, 1988). Interestingly, neither SAT sub-score was found to be statistically significant with negligible effects for Verbal SAT (F = 0.233; p = 0.792; f = 0.06) and Math SAT (F = 0.135; p = 0.874; f = 0.04). For the study approaches, the clusters differed with medium effects on both the deep approach (F = 4.190; p = 0.017; f = 0.22) and the surface approach (F = 7.315; p < 0.001; f = 0.27). Post hoc comparisons using the Tukey test indicate that the significant difference is Cluster 1 is higher on the surface approach than the other two clusters and Cluster 2 is higher on the deep approach than Cluster 1. On the final exam metric, the clusters were also different with a near medium effect (F = 3.663; p = 0.028; f = 0.21). Post hoc analysis describes the significant difference as Cluster 2 higher than Cluster 1. An ANCOVA analysis controlling for SAT sub-scores on the final exam measure indicated similar results (F = 3.881; f = 0.24) as the original ANOVA analysis.
Variables | Cluster 1 | Cluster 2 | Cluster 3 |
---|---|---|---|
Average (St. Dev.) | |||
SAT math | 566 (62) | 561 (71) | 560 (57) |
SAT verbal | 557 (70) | 566 (81) | 558 (71) |
Deep approach | 30.4 (6.5) | 33.7 (6.4) | 31.8 (5.8) |
Surface approach | 25.7 (5.9) | 22.1 (6.1) | 22.4 (5.6) |
Final exam | 43.8 (13.3) | 51.1 (15.2) | 46.8 (14.0) |
Thus, it appears that Cluster 2, which comprises roughly one-third of the sample, had higher scores on average on the final exam measure. This suggests the study approaches described as reviewing the textbook and practicing problems leads to increased academic performance in the course. Not surprisingly, Cluster 1, which indicated predominately not studying, performed worse. That Cluster 2 scored higher on the deep approach and Cluster 1 on the surface approach lends external validity to the qualitative difference between these two groups. Cluster 3's performance on the final exam is interesting as it was comparable to Cluster 1. The study efforts of Cluster 3 are more concentrated on the online homework. It is hypothesized that these students perceived the completion of the required online homework as suitable preparation for the exams. The central feature of the hypothesis is the emphasis on perception. Since the online homework was a mandatory part of the class it is likely that the strong majority of students completed it, however the students in Cluster 3 may have perceived it as satisfactory preparation whereas students in Cluster 2 believed that additional preparation was necessary. Thus, Cluster 1 may be described as knowingly not studying, Cluster 3 as believing the required course components constitute satisfactory studying and Cluster 2 studied in addition to the required course components by relying on the textbook and practicing problems.
The relationship of the study habits measured by ESM to a measure of academic performance serves as support for external validity of the data collected. The finding that students who study more regularly perform better on the cumulative final exam may not be surprising. However, the findings that approximately one-third of the sample study regularly, which matches the baseline observed in Fig. 2, is of importance as it suggests that there is ample ground for promoting effective study habits. That the students who study regularly are also not distinguishable from the other groups based on SAT scores also partially rules out the competing explanation that these students were more academically prepared prior to the semester. Another possible explanation for differences in academic performance may include differences between clusters in student motivation to succeed in the course; in particular, it is plausible that differences in motivation may manifest themselves in more frequent studying.
The end result created 18 categories from the responses across the set of four inquiries. Note these categories were created independently of the codes described in Table 3. A sizable advantage of the lexical analysis technique is the ability to demonstrate the categories and interrelations between each category in a web diagram. Web diagrams were created for each exam (Fig. 3 through 6) focusing only on categories with at least five responses. The web diagram represents each category with a node, and the size of the node is proportional to the frequency of the category. The frequency of each category is indicated in parenthesis inside each node. Nodes are connected with lines that indicate the extent the connected categories were mentioned together in a response. The type of line indicates the extent the categories are shared as a proportion of the smaller node. A solid line indicates that 60% or more of the responses that were categorized by the smaller node were also present in the category in the larger node. A long dash line indicates 40% to 59% agreement, a square dotted line indicates 20% to 39% agreement and no line indicates below 20% agreement. Reviewing the web diagrams can provide insight into changes in study patterns that occurred throughout the term. For context in interpreting trends in the web diagrams, the relevant topics from each exam are shown in Table 6.
Exam | Content |
---|---|
Exam 1 | Properties of light, electron configurations, periodic trends |
Exam 2 | Lewis structures, molecular shapes, polarity |
Exam 3 | Gas laws, thermodynamics |
Final exam | Intermolecular forces, colligative properties, cumulative exam included prior content |
In the Exam 1 (Fig. 3) diagram, notes, previous tests, textbook and homework are the most prominent, with PowerPoint (PPT) slides also mentioned. The links show moderate overlap among these five categories, though notably no significant overlap was found between previous tests and PowerPoint or previous tests and textbook. In the Exam 2 (Fig. 4) diagram the study pattern is more concentrated on notes, previous tests and textbook with moderate overlap among almost all of the categories. In Exam 3 (Fig. 5) the textbook is reduced in prominence and the studying was more focused on previous tests; also the relations among nodes are generally weaker than in Exam 2 indicating less reliance on multiple study approaches. In preparing for the cumulative Final Exam (Fig. 6) the use of the textbook has returned to prominence along with notes and previous tests, similar to Exam 2. This diagram is also the most interconnected web suggesting a stronger reliance on multiple studying techniques, possibly owing to the cumulative nature of this exam.
Looking for changes across study patterns, one clear trend is the diminished role of studying homework in preparation for the exams. In Exam 1, homework was among the most prominent nodes, whereas in each subsequent exam it is a minor node. This may describe students' perceiving a lack of relevance of the homework assignments in exam preparation after the first exam. For context, the online homework was always due one to two days before each of these four text message inquiries so that it was likely students were working on the assignments in the time frame indicated. Students could also review the homework assignments after the due date. Incidentally, after the semester had completed, the instructors at the setting discussed deliberately including one or two questions modified from the homework assignment in each of the exams to emphasize to students the importance of understanding the process of problem solving in the homework over simply arriving at the correct answer. By making this change it is possible that students may benefit more from engaging in the homework which would be reflected in their study habits and related to their academic performance.
Another trend among the web diagrams is the diminished role of the textbook and notes in the Exam 3 diagram. Exam 3 strongly relied on math content (see Table 6) differing from the preceding exams. Students may respond to this by studying the textbook and notes less and focusing more on the instructor provided materials in the PowerPoint slides and previous tests. Other explanations are also possible such as time constraints related to other courses giving exams, the perceived challenge of the practice tests that were posted taking up more student time or students finding the textbook less helpful in this content.
Returning to the research question, there appears to be considerable evidence of changing study habits among a common group of students over the course of the semester. The changing role of homework, textbook, notes and the use of multiple study techniques suggest that student study habits differ across the exams. The changes may be for many reasons including students responding to the perceived effectiveness of study techniques for each exam, the perceived nature of the content on each exam or the quality of study materials available, or competing interests for students' time. The changes in the nature of links also indicates that the variety of techniques used by students changes over the course of the semester and are amplified when taking a final exam, possibly as a result of the cumulative nature of the exam.
Next, the study provided evidence that study habits are related to academic performance in the course, notably by students using study habits that are in addition to the mandated course requirements. In this study, use of the textbook was most prominent as the additional study habit. A direct instructional implication that results from this study is the potential benefit of discussing with students the need to study beyond mandatory course components. In the current study one-third of the students described mandatory course components as their principal means for studying and these students performed comparably to students who did not report studying. Future research could have instructors discuss with students the results shown here to students and measure the impact on student study habits or academic performance.
One of the most interesting areas of future work may be an investigation on the impact of instructional techniques to impact study habits. Indeed one of the more surprising outcomes of this study was the infrequent mention of group work or studying with friends (Table 3) as it seemed possible that the weekly group work during the peer-led meetings would facilitate greater use of study groups outside of class. To investigate this area further, the impact of incorporating reform pedagogy or training sessions on how to study can be investigated in either a repeated measures or quasi-experimental design using text messages to measure students' study habits. Such a study may inform causal mechanisms behind evidence-based instructional practices. For example proponents of cooperative learning have indicated social constructivism as a potential explanation for improved learning outcomes that have been observed (Muthyala and Wei, 2013). The causal mechanism for learning would be that students' social processes within group-work have facilitated their conceptual understanding. An alternative causal mechanism however is that students engaged in cooperative learning may become dissatisfied with their own progress in comparison to their peers and as a result study more. The plausibility of the alternative explanation is supported by the time available outside of class relative to in class and the observed academic benefits of study habits herein. An investigation into the impact of evidence-based instructional practices on study habits can then support or dissuade the alternative explanation proposed.
This study has also shown that students' study habits can change over the course of the semester. This finding has relevance for work that relies on a single measure of particular study habits (as opposed to more general study approaches) extrapolated to describe students' habits throughout the term. It also provides some support to the expectation that instructor actions can influence student study habits. Relating changes in study habits to measures of reflective action (e.g. metacognition) or interviewing students regarding their study actions prior to each test may offer additional support for this contention. Additionally future research that investigates how changes in study habits relate to academic performance is warranted, as whether consistent or adaptable study habits are more beneficial remains an open question.
Finally, ESM has the flexibility to potentially support a diverse range of instructional strategies. For example, the action of messaging students outside of class can, by itself, serve as an instructional intervention. Instructors can use text messages to direct students toward online resources, set-up peer study groups or remind students of deadlines in a timely fashion. Additionally, the messages can be tailored for individual students or small groups; for example messages can notify a group of students who haven't completed an online homework assignment of an upcoming due date or inform a student who has struggled that the student's recent test score shows an improvement over past performance. Early research in a wide range of educational settings has shown that such tailored messages have a strong potential for producing positive gains (Dynarski, 2015). This approach may offer a non-intrusive way to show faculty concern for student performance which, when missing, has been cited as a factor in student attrition from the STEM disciplines, particularly among minority students (Tsui, 2007; Museus et al., 2011).
Footnote |
† Analyses presented later in this manuscript will rely on a subset of the sample based on frequencies of responses to text messages. The correlation between frequency of responses and each of the measures in Table 2 were found to be weak, with r = 0.16 for response rate to Math SAT or Verbal SAT and |r| < 0.14 for the other measures, indicating subsets generally continue with minor departures from the population. |
This journal is © The Royal Society of Chemistry 2015 |