College Students’ Conceptions of Learning of and Approaches to Learning Computer Science

Abstract

The purpose of this research was to examine college students’ conceptions of learning computer science and approaches to learning computer science and to examine the relationships among these two important constructs and possible moderating factors. Student data (N = 193) were collected using the conceptions of learning computer science and the approaches to learning computer science surveys at one public research institution in the southeastern United States. Data were analyzed with descriptive statistics, Confirmatory Factor Analysis models, internal consistency reliability, Pearson correlations, stepwise multiple regression models, and Multivariate Analysis of Variance models. The results suggest that college students most favorably employ a deep strategy approach for learning computer science in which prior knowledge is activated and meaningful learning strategies are used. College students appear to be more extrinsically motivated to learn computer science than intrinsically. Higher level learning conceptions are associated with a deep strategy approach to learning (e.g., Seeing in a new way) whereas low-level conceptions are associated with a surface strategy (e.g., Memorizing) approach to learning. Male college students have slightly higher conceptions of programming than their female counterparts. The findings are discussed and both limitations and delimitations of the study are enumerated.

Keywords

Computer science conceptions of learning approaches to learning individual differences

Introduction

Computer science has been acknowledged as a critical discipline for the future of the information economy. Two presidents of the United States (former president Obama and president Trump) have formally underscored the importance of the field to our future economy and society and called for additional funding to support the development of computer science education both in K-12 schools and institutions of higher education (Obama White House, 2016; White House, 2017). Agencies such as the National Science Foundation and the Department of Education have established competitive grant programs to increase access and quality to computer science learning experiences. Private companies have also committed substantial resources to computer science education, including Google, Facebook, and Amazon (Wan, 2017). Professional associations such as the International Society for Technology in Education, the Association for Computing Machinery, and the Computer Science Teachers Association have also contributed to the cause by developing standards and definitions used in educational practice. Nonprofit organizations such as Code.org provide resources for students and teachers directed at spreading computer science education learning experiences.

As a consequence of these synergistic efforts, enrollments in computer science undergraduate degree programs have risen markedly in the United States since 2005 according to recent reports (National Academies, 2018). However, even with these increased enrollments in computer science degree programs, participation in these degree programs has had a low representation of minorities and women (National Academies, 2018). Furthermore, many students intending to major in computer science do not persist with the major to degree completion. With the increased emphasis in computer science education, educators and researchers are now turning to rigorous and relevant research methods to illuminate the complexity of computer science education among K-12 and higher education students. For example, introductory computer programming courses are often a major hurdle for students entering the major since computer programming is a requisite skill for the major. Evidence suggests that introductory programming courses in institutions of higher education have a pass rate of approximately 68%, which suggests that about 32% of the students enrolling introductory programming courses (also known as CS1) are not passing the course on the first attempt (Watson & Li, 2014). Given the criticality and crisis surrounding CS1, extensive research work has been conducted on various educational issues such as gender differences, pair programming, blended learning, block programming, and gamification. The literature on intermediate programming courses such as CS2 (second programming course) and CS3 (data structures) has been quite limited. Thus, relatively few studies have examined how perceptions of students enrolled in introductory programming courses vary from that of intermediate programming courses.

An area of importance to computer science education research is a fundamental understanding of students’ conceptions of learning computer science (COLCS). What do we know about the nature of learning computer science among undergraduate students? Students conceptions of learning have been connected to many other important constructs in education contexts, including approaches to learning (Lee, Johanson, & Tsai, 2008), study habits (van Rossum & Schenk, 1984), epistemological beliefs (Tsai, Ho, Liang, & Lin, 2011), self-efficacy (Tsai et al., 2011), and academic achievement (McCracken, 2001). Equally important, a firm grasp of the approaches undergraduate students take to learning computer science could shed light on successful instructional strategies to promote computer science learning. What approaches to undergraduate students take to learn computer science? More importantly, what are the relationships among student COLCS, and the approaches they take to learning computer science? This study explores these ideas with the goal of better understanding of undergraduate students in computer science. Such information could lead to improved instructional strategies for all computer science students and a deeper understanding of the nature of learning computer science.

Supporting Literature

Several previous studies have attempted to shed light on student COLCS, and on the approaches to learning computer science (ALCS). However, we were only able to identify one study that simultaneously examined both concepts in an effort to explore meaningful relationships between conceptions and learning approaches. Thus, we modeled our study after the work of Liang, Su, and Tsai (2015). We review previous studies to show the unique contributions of this study to the knowledge base.

Conceptions of Learning Computer Science

Conceptions of learning research refers to students’ views about their learning experiences and their preferred ways of undertaking the learning process in a subject domain like computer science. Säljö (1979) was the early pioneer in this body of research who started a series of relevant research studies to identify college students’ conceptions of learning. Using interviews as a research method, Säljö (1979) found the following five categories: (a) increase of knowledge, (b) memorizing, (c) acquisition of facts or procedures that can be retained and/or utilized in practice, (d) abstraction of meaning, and (e) interpretative process aimed at the understanding of reality. In this research, we are concerned with student learning conceptions of computer science. Since then, several other researchers have attempted to examine the domain of computer science.

Bruce et al. (2004) investigated students’ early experiences of learning programming. They conducted semistructured interviews of students who had either recently completed or enrolled in an introductory computer programming course. They interviewed 13 students from first-year Java programming course of a Bachelor of Information Technology degree. Their analysis revealed five categories of conceptions experienced by the students: (a) following—getting through the programming assignments by following the given instructions, (b) coding—learning the syntax of the language, (c) understanding and integrating—learning new concepts and integrating with known concepts and tasks to be completed, (d) problem-solving—doing what it takes to solve the problem at hand to complete the tasks, and (e) participating or enculturation—seeing differences how they think too that of a programmer and understanding what it takes to be part of programmer community.

Thuné and Eckerdal (2009) studied novice students’ conceptions of computer programming by interviewing students and recording their expressions of computer programming. They interviewed first-year engineering students enrolled in a computer programming course at a university in Sweden. They interviewed the students at the end of the semester. They analyzed the transcribed data to identify the categorical descriptions of how students’ see computer programming. Their analysis revealed five categories—computer programming as (a) writing text, (b) describing actions, (c) producing tools for everyday life use, (d) solving problems, and (e) empowering skill in various contexts. They applied variation theory and patterns of variation to analyze the categorical findings. Based on the analysis, they make recommendations for instructors on how to improve student experiences using the categorical dimensions they identified.

In a dissertation study, Hewner (2012) interviewed 37 students and advisors to investigate student conceptions of the computer science discipline. He analyzed the transcripts with a grounded theory approach to identify conceptions. The study findings revealed three main conceptions of computer science held by students: (a) computer science is mathematical study of algorithms, (b) computer science is mostly about programming, and (c) computer science has many subfields with focus on programming as well nonprogramming. Hewner (2012) connected student learning conceptions of computer science to student educational decisions and found a weak relationship between the two constructs. However, receiving poor grades in computer science courses did appear to provoke switching majors.

Liang et al. (2015) explored the relationship between two surveys—COLCS and ALCS. Both surveys were completed by 421 college students majoring in computer science in Taiwan. The authors performed principal component analysis, Pearson correlations, and stepwise multiple regression analyses. Their analyses showed that both surveys were sufficiently reliable to assess students’ learning conceptions, and ALCS. The COLCS included seven measures of student learning conceptions in computer science: Memorizing, Testing, Calculating and practicing, Programming, Increasing one’s knowledge, Application and understanding, and Seeing in a new way. These dimensions were the COLCS identified in the study.

Approaches to Learning Computer Science

Approaches to learning are students’ ways of processing various tasks in a situated learning environment (Biggs, 1987). This situated learning environment is this study is the domain of undergraduate computer science. The learning process is complex and composed of students’ motives and strategies for learning, and each motive and strategy combination defines an approach to learning (Biggs, 1987). With regard to approaches to learning, deep approach and surface approach are the two distinct categories. Using deep approach, the learner intends to understand the ideas for the learner self and is correlated with intrinsic motivation. The learner aims at transforming knowledge into meaningful use. The learner also actively relates the idea to previous knowledge and experience, and looks for patterns and underlying rules (Entwistle, 1998). Using surface approach, nevertheless, the learner aims at reproducing the information from the learning material in order to cope with course or test requirements. The learner is driven by extraneous motivation and studies without reflecting on the purpose, strategy, and achievement. The main strategy the learner uses would be memorizing facts and procedures without relating them to prior knowledge (Entwistle, 1998).

In the field of computer science, approaches to learning can be powerful predictors of achievement, particularly in computer programming courses (de Raadt et al., 2005). In this study, students employing a deep approach or deep motive have significant positive correlations with student achievement, whereas students employing a surface approach or surface motive had a significant negative correlation (de Raadt et al., 2005). The correlations between approaches to learning were stronger than other measures such as prior knowledge of programming or a general measure of cognition. However, we have evidence that computer science students do not always adopt the deep approach to learning (Malakolunthu & Joshua, 2012). Furthermore, the approach to learning computer science can be mediated by the multiple factors, like the context of the learning environment, students’ value and motivation, the professional capacity of the computer science instructors, the assessment methods, and the instructional rigor of the computer science curriculum (Malakolunthu & Joshua, 2012).

Forte and Guzdial (2005) tried to tailor a CS1 course to accommodate the interest and background of a variety of non-CS major students. Through designing tasks that were targeting the non-CS major students, they could relate computer programming with their prior knowledge in other fields. The results indicated a deep approach of learning for these non-CS major students by scaffolding learners to meet their needs. Notably, students’ pass rate and engagement significantly increased (Forte & Guzdial, 2005). In another study, Wishart (2005) used the Two-Factor Study Process Questionnaire (Biggs, 2001) to measure undergraduate students’ learning approach and connect them to learning styles. The study explicitly compared computer science students with information science students. Results of this study showed that computer science students were more likely to prefer problem-solving and adopt a deep approach to learning (Wishart, 2005).

Revisiting the study by Liang et al. (2015), the survey results from the ALCS produced four distinct factors: Surface motive, Surface strategy, Deep motive, and Deep strategy. These results were examined in relation to the seven factors noted from the COLCS using Pearson correlations and stepwise multiple regression analysis. The results showed that Memorizing, Testing, and Calculating and practicing had significant positive correlations with the Surface strategy, whereas Programming, Increasing one’s knowledge, application and understanding, and Seeing in a new way had significant positive correlations with Deep strategy. Their results show that there are deep relationships between students’ COLCS and students’ ALCS.

Purpose Statement

The purpose of this study is threefold: (a) to characterize college student’s COLCS and ALCS, (b) to explore the relationships between COLCS and ALCS, and (c) to explore the differences among college student’s COLCS and ALCS on gender and minority status. Most computer science instructors might be familiar seldom implement their classroom strategies to be gender and minorities inclusive as they may not be aware of how to make changes to be inclusive. Thus, studying students’ learning conceptions differences from gender and minorities perspectives can provide insightful information for instructors to prepare classroom materials that are gender and minorities inclusive. Inspired by a similar study conducted with Taiwanese college students (Liang et al., 2015), our study extends this work with college students from the United States and rigorously examines the survey measures using confirmatory factor analysis (CFA) models. Our guiding research questions are as follows: (a) what conceptions do college students have about learning computer science, (b) what approaches do college students take to learning computer science, and (c) what are the relationships among these two constructs, and (d) how do these constructs differ by minority status and gender?

Method

Institution

All data were collected from a public comprehensive university in the southeastern United States serving both traditional and nontraditional students. The regionally accredited institution serves approximately 16,000 students in primarily bachelor’s degree programs and select graduate degree programs. The institution has a diverse academic unit in computing disciplines with more than 15 full-time faculty members serving approximately 750 students. The schools offer a bachelor’s degree in computer and information sciences with four Accreditation Board for Engineering and Technology accredited tracks: (a) computer science, (b) information science, (c) information technology, and (d) information systems. Multiple computer programming courses are embedded in all four of the degree tracks.

Participants

One-hundred ninety-three of these computer and information science students (N = 193) participated in the research study with a typical computing discipline gender distribution of 78% males and only 22% females. Approximately 83% of the participants were full-time students with an average age of 22.27 (SD = 6.65). The students were at different points in their academic majors with 32% freshman, 22% sophomore, and 42% juniors (few seniors were in the sample). Seventy-two percent of the sample identified as White, 12% as Black/African American, and 11% as Hispanic/Latino (participants could choose multiple race/ethnicities on the survey). Most students reported the course in which we collected data as their first college computer science course at 47%, while 26% reported having taken one to two courses, and 17% reported having taken three to four computer science courses. Notably, 37% of the student participants had taken a high school computer science course, and approximately 10% of the participants were already working in the field in a full-time or internship position.

Instruments and Measures

Background and demographics

The background survey had nine separate items and collected basic demographic information about each student participant, including their gender, age, ethnicity, student classification (e.g., freshman), number of previous college computer science courses, high school programming experience, employment status, and degree program.

Conceptions of Learning Computer Science

The survey was originally designed and developed by Lee et al. (2008) to measure high school seniors’ conceptions of learning science. The survey was renamed as the COLCS and was adapted later to assess college students majoring in computer science related disciplines (Liang et al., 2015) using a 7-point Likert-type scale, which ranges from strongly disagree (1) to strongly agree (7). The modified survey measured seven constructs on a sample of N = 421 college student participants from Taiwan using principal component analysis and a varimax rotation. The seven-factor model accounted for approximately 75% of the variability in these data and was organized into the following factors (Liang et al., 2015):

“Memorizing” (three items): Learning computer science is viewed as memorizing what teacher lectures in the class, related definitions, rules, and code, as well as important concepts described in the textbooks that can help answer teacher’s questions.

“Testing” (four items): Learning computer science means getting familiar with test materials, passing the exams, and getting high scores in the tests.

“Calculating and practicing” (three items): Computer science is learned through practicing solving the coding problems and calculations and repeating using the program and formula.

“Programming” (four items): Learning computer science is considered as learning how to program to improve performance in the course.

“Increasing one’s knowledge” (3 items): Learning computer science is regarded as acquiring computer science knowledge that student did not know before.

“Application and understanding” (four items): Learning computer science aims at understanding relevant knowledge and applying it to solve problems.

“Seeing in a new way” (five items): Learning computer science is considered as using a new viewpoint, changing perspective, or finding a more reasonable explanation of the phenomena related to computer science and topics surrounding our lives.

All of the reliability coefficients in their model were above the .70 social science threshold with factor loadings all above .50 for the 26-item instrument (Liang et al., 2015). In this study, we maintained the same 26 items with one key difference: we used a 5-point Likert-type scale for the measures (strongly disagree, disagree, neither agree nor disagree, agree, and strongly agree). As surveys were conducted using paper and pencil format, we used 5-point scale instead of 7-point scale as longer scales tend to yield worse data quality (Yan, Keusch, & He, 2018).

Approaches to learning computer science

The original version of the ALCS was designed by Kember, Biggs, and Leung (2004) and later revised by Liang, Lee, and Tsai (2010) to measure student approaches to learning science. The ALCS was substantially revised for computer science contexts by Liang et al. (2015) in a study with N = 421 college student participants from Taiwan using principal component analysis and a varimax rotation. The ALCS used a 7-point scale from Strongly Disagree to Strongly Agree. The analysis of the ALCS revealed that the students’ responses to the survey were grouped into four distinct factors (Liang et al., 2015):

“Surface motive” (three items): Learning computer science is driven by external motivations like passing the exam, meeting the degree requirement, getting a well-paid job, or pleasing parents and other family members.

“Surface strategy” (four items): Learning computer science is achieved by memorizing the most important concepts, rules, and codes to achieve high scores in exams.

“Deep motive” (six items): Learning computer science is driven by students’ interest and curiosity as well as internalized external motivation

“Deep strategy” (four items): Learning computer science is achieved by higher order thinking, like attaining coherent understanding, making connections of knowledge and skills, and applying knowledge into use.

The model explained approximately 69.65% of the variability in the data with all Cronbach’s αs above the .70 mark (Liang et al., 2015). Again, we maintained the 17-items and used a 5-point scale for the measurement system.

Procedures

All student participants were recruited from one metropolitan, comprehensive public university in the southeastern United States. All instructors at the institution teaching computer programming courses were sent an e-mail early in Fall 2018 academic semester to request permission for a member of the research team to collect data from their courses at a designated date and time approved by the instructor. Out of the 14 courses invited to participate, 9 agreed and were scheduled to collect these data. Titles of these courses ranged from introduction to object-oriented programming to computer science one to data structures. Instructors were encouraged to offer a small extra credit incentive for the students to participate in the research program. Upon attending a scheduled class, a member of the research team briefly described the goals of the research program while distributing the survey materials and informed consent documents to the student participants. As students completed the task, they returned the paper survey to the researcher. This process was repeated for each class. Next, these paper survey data were carefully ported to a digital version of the survey by members of the research team to prepare these data for subsequent analyses.

Data Analysis

This study employed both descriptive and inferential statistical methods using SPSS version 25 and Mplus version 8.0. We assessed the data using descriptive statistics, including the mean, standard deviation, skewness, and kurtosis for each item. Since our study was inspired by a similar study conducted with Taiwanese college students (Liang et al., 2015) using established measurements, we chose to employ CFA to test the factor structure of the COLCS and ALCS survey tools for these data collected from students within the United States. Several goodness-of-fit indices were used to assess the fitness of the research model to the data: χ², root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), comparative fit index (CFI), and Tucker–Lewis Index (TLI). After assessing the measurement quality of the models, data were subjected to Pearson correlations and stepwise multiple regression models to examine the relationships among the survey constructs. Finally, we conducted multivariate analysis of variance (MANOVA) on each of the factors in relation to minority status and gender.

Results

We present our results by each of the measurement systems: COLCS and ALCS to first characterize the student participant sample and evaluate the fit of the data on the model using CFA based on the theoretical factor model (Liang et al., 2015). After, we examine the relationships among the factors from the two surveys using Pearson correlations and stepwise multiple regression models. We conclude by conducting MANOVA on each of the factors from the COLCS and ALCS surveys with minority status and gender serving as between subject conditions. We explore these variables to discern if conceptions of learning and ALCS vary by minority status or gender since these two individual differences are areas in which we seek to broaden participation.

Conceptions of Learning Computer Science

Table 1 provides the descriptive statistics and model fit data for the COLCS. The descriptive statistic results indicate acceptable normality from the majority of the items. Only four items had Kurtosis outside of −2 and 2 range. In this case, we considered them as continuous variables and use Maximum Likelihood as the estimation method. Chi-square test was used to indicate the amount of difference between expected and observed covariance matrices (Bentler & Bonett, 1980). The model result did not indicate good exact fit, χ² (df = 278, N = 193) = 493.82, p < .001. Thus, we examined the close fit. The CFI and TLI were computed as well to assess model fit, with values of CFI and TLI greater than 0.90 indicating good model fit (Bentler, 1989; Bentler & Bonett, 1980). We also examined the RMSEA and SRMR. Acceptable model fit is indicated by an RMSEA value of 0.06 or less and an SRMR value of 0.10 or less (Hu & Bentler, 1999). The fit indices show CFI = 0.87, TLI = 0.84, RMSEA = .063, with 90% confidence intervals for RMSEA: [0.05, 0.07], p value for RMSEA ≤ .05 = 0.009, SRMR = 0.068. This model does demonstrate an acceptable fit to the data.

Table 1.

Descriptive Statistics and Factor Loadings for Each Item on the COLCS.

Item/Factor (Cronbach’s α)	Mean	SD	Skewness	Kurtosis	Factor loading	p
Memorizing (α = .73)	3.08	0.86	−0.28	−0.18
Learning computer science means memorizing the important concepts found in a computer science textbook	3.11	1.09	−0.18	−1.06	.73	<.01
Learning computer science means memorizing the proper nouns found in a computer science textbook that can help solve the teacher’s questions	2.60	1.09	0.29	−0.81	.79	<.01
Learning computer science means remembering what the teacher lectures about in the class	3.54	1.03	−0.77	0.12	.57	<.01
Testing (α = .69)	2.35	0.74	0.20	−0.30
Learning computer science means getting high scores on examinations	2.76	1.11	0.01	−0.80	.50	<.01
The major purpose of learning computer science is to get more familiar with test materials	2.13	1.01	0.65	−0.26	.82	<.01
I learn computer science mainly because of the tests.	1.96	0.97	0.83	−0.12	.64	<.01
There is a close relationship between learning computer science and taking tests	2.58	1.02	0.27	−0.50	.44	<.01
Calculating and practicing (α = .81)	4.16	0.74	−1.29	2.65
Learning computer science means the process of a series of calculation and solving problem	4.21	0.83	−1.50	3.42	.69	<.01
Learning computer science means constantly practicing calculation and solving problem	4.06	0.94	−1.05	1.04	.81	<.01
The way to learn computer science well is to constantly practice calculations and problem-solving	4.20	0.85	−1.12	1.41	.81	<.01
Programming (α = .78)	4.47	0.52	−1.02	1.78
Learning computer science means constantly practicing the skills to design computer programs	4.34	0.74	−0.79	−0.22	.62	<.01
Learning programming will help me improve my performance in computer science courses	4.59	0.56	−1.31	2.61	.69	<.01
Learning computer science means learning computer programming skills	4.43	0.69	−1.18	1.51	.71	<.01
The way to learn computer science well is to constantly practice and get more familiar with programming skills	4.52	0.66	−1.71	4.60	.74	<.01
Increasing one’s knowledge (α = .56)	3.97	0.63	−0.43	0.04
Learning computer science means the acquisition of knowledge about computer science	4.30	0.74	−1.00	1.47	.63	<.01
Learning computer science means acquiring knowledge that I did not know before	4.26	0.78	−1.01	1.18	.52	<.01
I am learning computer science when the teacher tells me scientific facts that I did not know before	3.34	1.05	−0.31	−0.42	.53	<.01
Application and understanding (α = .77)	3.83	0.74	−0.74	0.82
Learning science means acquiring knowledge and skills to enhance the quality of our lives	3.81	1.00	−0.58	−0.19	.69	<.01
Learning computer science means learning how to apply the knowledge and skills I already know to unknown problems	4.10	0.93	−1.10	1.04	.66	<.01
Learning computer science means understanding the connection between scientific concepts	3.75	0.91	−0.52	0.08	.68	<.01
Learning computer science means an understanding of some problems or phenomena that cannot be solved before	3.64	1.01	−0.51	−0.15	.68	<.01
Seeing in a new way (α = .76)	3.93	0.62	−0.75	2.30
Learning computer science means using a new viewpoint to view the phenomena or topics related to computer science	4.01	0.80	−1.05	2.08	.64	<.01
Learning computer science means changing my way of viewing the phenomena and topics related to computer science	3.82	0.88	−0.74	0.59	.58	<.01
Learning computer science means finding a better way to view the phenomena or topics relating to computer science	4.02	0.85	−0.90	1.19	.82	<.01
I can learn more ways of thinking about the phenomena or topics relating to computer science by learning it	4.14	0.79	−0.96	1.26	.64	<.01
Learning computer science means finding a more reasonable way to explain the topics surrounded our lives	3.64	1.00	−0.38	−0.42	.51	<.01

Note. COLCS = conceptions of learning computer science.

As can be gleaned in Table 1, the highest factor mean was for the Programming construct at M = 4.47, whereas the lowest factor is for the Testing construct at M = 2.35. Two of the constructs did not result in satisfactory internal consistency reliability as calculated by Cronbach’s α: Testing (α = .69) and Increasing one’s knowledge (α = .56). The estimation of the loadings of the items on the factors shows a high correlation between the factors and the corresponding items ranging from .44 to .82. About half of the items’ factor loadings are above or close to .70, which means that the factors can explain at least 50% of the variance of these items. Only the factor loading of Item 7 on the factor Testing is less than .50.

Approaches to Learning Computer Science

Table 2 provides the descriptive statistics and model fit data for the ALCS. Descriptive statistics also indicates good normality of the items. Only two items had kurtosis outside of the −2 and 2 range. We also used Maximum Likelihood as the estimation method. We conducted a CFA to validate the factors of the ALCS survey. The model results did not indicate good exact fit from the data, χ² (df = 113, N = 193) = 225.76, p < .001. The close fit results show CFI = 0.91, TLI = 0.89, RMSEA = .072, with 90% confidence intervals for RMSEA: [0.06, 0.09], p value for RMSEA ≤ .05 = 0.005, SRMR = 0.073. This model does demonstrate an acceptable fit to the data. The standardized factor loading of each item is included in Table 5. The factor loadings have a wider range, from .23 to .91. Majority of the items have factor loading higher than .50. Half of the items’ factor loadings are higher than .70, which means that the corresponding factor can explain over 50% of the variance of these items. Third item under surface motive and first two items under the surface strategy categories need to be looked into due to the factor loadings less than .5. Both the Surface motive and Surface strategy factors demonstrated less than desirable internal consistency reliability coefficients. The highest rated factor was the Deep strategy construct at M = 4.06, while the lowest rated factor was Surface strategy at M = 2.68.

Table 2.

Descriptive Statistics and Factor Loadings for Each Item on the ALCS.

Item/factor (Cronbach’s α)	Mean	SD	Skewness	Kurtosis	Factor loading	p
Surface motive (α = .63)	3.98	0.75	−0.87	0.99
Whether I like it or not, I can see that doing well in school is a good way to get a well-paid job	4.25	0.85	−1.26	1.78	.56	<.01
I want to get a good achievement in computer science subject so that I can get a better job in the future	4.31	0.85	−1.39	2.06	.89	<.01
I want to do well in computer science subjects so I can please my family and the teacher	3.38	1.23	−0.43	−0.69	.46	<.01
Surface strategy (α = .61)	2.68	0.79	0.15	−0.54
I find that studying each topic in depth is not helpful or necessary when I am learning computer science. There are too many examinations to pass and too many subjects to be learned	2.69	1.13	0.26	−0.75	.31	<.01
I learn computer science by rote, going over and over them until I know them by heart	3.18	1.12	−0.27	−0.74	.23	<.01
I find the best way to pass computer science examinations is to try to remember the answers to likely questions	2.49	1.18	0.32	−0.96	.74	<.01
I find that memorizing the most important content of computer science-related subject instead of understanding the content, so that I can get high scores in most examinations	2.36	1.21	0.47	−0.98	.82	<.01
Deep motive (α = .88)	3.81	0.86	−1.03	1.29
I find that at times studying computer science makes me feel really happy and satisfied	3.86	1.08	−1.01	0.50	.77	<.01
I feel that computer science topics can be highly interesting once I get into them	4.32	0.85	−1.63	3.48	.81	<.01
I work hard at studying computer science because I find the material interesting	3.89	1.06	−0.79	0.01	.87	<.01
I always greatly look forward to go to computer science class	3.49	1.11	−0.49	−0.43	.75	<.01
I find that I continually go over my computer science class work in my mind even whenever I am not in computer science class	3.52	1.21	−0.49	−0.71	.65	<.01
I like to work on computer science topics by myself so that I can form my own conclusions and feel satisfied	3.74	1.17	−0.64	−0.45	.64	<.01
Deep strategy (α = .80)	4.06	0.67	−0.79	0.94
I try to find the relationship between the contents of what I have learned in computer science subjects	4.16	0.81	−0.88	0.82	.79	<.01
I try to relate new material to what I already know about the topic when I am studying computer science	4.21	0.81	−1.04	1.21	.91	<.01
I try to understand the meaning of the contents I have read in computer science textbooks	3.95	0.96	−0.99	0.86	.59	<.01
I can ask myself possibly to understand the subject matter I have learned in computer science class	3.92	0.78	−0.32	0.00	.58	<.01

Note. ALCS = approaches to learning computer science.

Correlations Between Conceptions of Learning and Approaches to Learning Computer Science

We conducted Pearson correlations among the COLCS and ALCS factors to examine the nature of the relationships among conceptions of learning and ALCS. These results are presented in Table 3. As can be gleaned, the strongest and significant positive relationships were found between Calculating and practicing and Deep motive at r = .43, and Seeing in a new way and Deep motive at r = .43. Memorizing had significant and positive relationships with Surface motive and Surface strategy, and Testing had a negative relationship with both Deep motive and Deep strategy. In general, Surface strategy had a stronger, positive relationship with Memorizing and Testing while Deep strategy had stronger positive relationships with Increasing one’s knowledge, Application and understanding, and Seeing in a new way. While the Surface strategy had a significant negative relationship with Programming, Deep strategy had a significant positive relationship with Programming.

Table 3.

Correlations Between COLCS and ALCS.

COLCS/ALCS factors	Surface motive	Surface strategy	Deep motive	Deep strategy
Memorizing	.225**	.328**	0.035	−0.046
Testing	.165*	.380**	−.258**	−.230**
Calculating and practicing	−0.056	−.151*	.431**	.251**
Programming	0.037	−.253**	.285**	.241**
Increasing one’s knowledge	.251**	−0.023	.330**	.248**
Application and understanding	0.087	−0.115	.366**	.301**
Seeing in a new way	0.08	−0.132	.426**	.409**

Note. COLCS = conceptions of learning computer science; ALCS = approaches to learning computer science.

*p < .01. **p < .001.

Stepwise Multiple Regression Analyses for Predicting Students’ Approaches to Learning Computer Science

We conducted a series of stepwise multiple regression analyses to predict students’ ALCS. The students’ conceptions (i.e., Memorizing, Testing, Calculating, Programming, Increasing one’s knowledge, Application and understanding, and Seeing in a new way) were used as predictors, and their learning approaches (i.e., Surface motive, Surface strategy, Deep motive, and Deep strategy) were the outcomes of the analyses. The results are shown in Table 4. It was discovered that Testing was a significant predictor of ALCS in the Surface strategy, Deep motive, and Deep strategy constructs. Notably, Testing was a positive predictor on Surface strategy, but served as a negative predictor for both Deep motive and Deep strategy. Seeing in a new way as a positive significant predictor for both Deep motive and Deep strategy. Again, the data in summary show that the Testing and Memorizing are positive predictors for a Surface strategy whereas negative predictors for Deep motive and Deep strategy.

Table 4.

Stepwise Regression Models for Predicting Approaches to Learning Computer Science.

Approaches to learning computer science	B	SE	β	t	R ²
Surface motive
Constant	2.56	0.36		7.09**	.30
Increasing one’s knowledge	0.24	0.09	0.20	2.69**
Memorizing	0.16	0.06	0.18	2.49*
Surface strategy
Constant	3.07	0.50		6.11**	.49
Testing	0.28	0.08	0.27	3.79**
Programming	−0.38	0.10	−0.25	−3.79
Memorizing	0.22	0.07	0.24	3.333*
Deep motive
Constant	0.87	0.47		1.85	.58
Calculating and practicing	0.35	0.08	0.30	4.49**
Seeing in a new way	0.30	0.10	0.22	3.04*
Testing	−0.25	0.07	−0.22	−3.50*
Increasing one’s knowledge	0.23	0.09	0.17	2.41
Deep strategy
Constant	2.78	0.34		8.17**	.45
Seeing in a new way	0.42	0.07	0.39	5.78**
Testing	−0.15	0.06	−0.17	−2.55

*p < .01. **p < .001.

Exploring Differences by Gender and Minority Status

As gender and minority status are key individual differences related to the goal of broadening participation in computer science, we sought to explore whether there were differences on students’ conceptions of learning and ALCS by both gender and minority status. To test for these differences, a MANOVA model was run for the COLCS and ALCS factors as separate models for both gender and minority status. Table 5 shows the results for the COLCS by gender across the seven constructs. As can be gleaned, the only factor to have a significant difference on gender was Programming with male students having significantly higher conceptions of this construct than their female counterparts. As for the ALCS, Table 6 shows that there are no differences among the four factors based on gender.

Table 5.

Descriptive Statistics and MANOVA Results by Gender for COLCS Survey Factors.

	Females			Males
Factor	Mean	N	SD	Mean	N	SD	F	p
Memorizing	3.20	41	0.87	3.04	150	0.86	1.03	.31
Testing	2.47	42	0.73	2.32	151	0.74	1.08	.30
Calculating and practicing	4.32	41	0.53	4.11	149	0.79	2.01	.16
Programming	4.32	42	0.53	4.51	151	0.51	4.45	.04
Increasing one’s knowledge	3.87	42	0.57	4.00	149	0.65	1.18	.28
Application and understanding	3.89	42	0.75	3.81	151	0.74	0.66	.42
Seeing in a new way	4.01	42	0.55	3.91	150	0.64	1.36	.25

Note. MANOVA = multivariate analysis of variance; COLCS = conceptions of learning computer science.

Table 6.

Descriptive Statistics and MANOVA Results by Gender for ALCS Survey Factors.

	Females			Males
Factor	Mean	N	SD	Mean	N	SD	F	p
Surface motive	3.89	42	0.70	4.01	150	0.76	0.74	.39
Surface strategy	2.70	42	0.78	2.67	149	0.79	0.03	.87
Deep motive	3.64	42	0.87	3.85	149	0.85	1.81	.18
Deep strategy	4.11	42	0.68	4.04	151	0.67	0.41	.52

Note. MANOVA = multivariate analysis of variance; ALCS = approaches to learning computer science.

The next series of MANOVA models examined the differences based on minority status as classified as White and non-White across the factors on the COLCS and ALCS surveys. As shown in Table 7, there are no discernable and statistically significant differences between White and non-White conditions across the seven measures of the COLCS survey. Table 8 shows the results from the MANOVA on minority status for the four ALCS survey factors. While none of the constructs were statistically significant based on minority status, it is worth noting that the Surface strategy construct is approaching significance at p = .06 with non-White students favoring the Surface strategy approach more than their White counterparts.

Table 7.

Descriptive Statistics and MANOVA Results by Minority Status for COLCS Survey Factors.

	Non-White			White
Factor	Mean	N	SD	Mean	N	SD	F	p
Memorizing	3.22	52	0.73	3.02	139	0.90	1.91	.17
Testing	2.50	54	0.67	2.30	139	0.76	2.79	.10
Calculating and practicing	4.12	52	0.88	4.17	138	0.69	0.32	.58
Programming	4.45	54	0.55	4.48	139	0.51	0.20	.66
Increasing one’s knowledge	3.87	53	0.74	4.00	138	0.58	1.92	.17
Application and understanding	3.89	54	0.83	3.80	139	0.70	0.06	.80
Seeing in a new way	3.93	54	0.75	3.93	138	0.57	0.19	.66

Note. MANOVA = multivariate analysis of variance; COLCS = conceptions of learning computer science.

Table 8.

Descriptive Statistics and MANOVA Results by Minority Status for ALCS Survey Factors.

	Non-White			White
Factor	Mean	N	SD	Mean	N	SD	F	p
Surface motive	3.86	54	0.62	4.03	138	0.79	1.40	.24
Surface strategy	2.87	53	0.77	2.61	138	0.78	3.46	.06
Deep motive	3.68	53	0.99	3.86	138	0.80	2.08	.15
Deep strategy	3.93	54	0.68	4.11	139	0.66	2.17	.14

Note. MANOVA = multivariate analysis of variance; ALCS = approaches to learning computer science.

Discussion

Interpretation of these results must be viewed in light of the limitations and delimitations of this study. First, the sample of student participants was from one university in the southeastern United States. Caution should be taken in generalizing these results to all undergraduate computer science students. Second, there were some measurement issues with the COLCS and ALCS survey factors with some items not loading well on the factors and the internal consistency reliability for some of the factors below the social science norm of .7 (Nunnally, 1978). Although the CFA showed a reasonable fit for both models, the data cannot be assumed to be without measurement problems. Furthermore, the original administration of the COLCS and ALCS survey used a 7-point scale, whereas we used a 5-point scale in this study. Third, we only collected quantitative survey data in this study as opposed to a mixed method design, which would have enabled us to collect qualitative data to triangulate our findings and potentially provide deeper explanations. Finally, as this is survey research, the honesty and accuracy of the student participants is also a factor since the students may have provided “socially acceptable” responses. With these considerations in mind, our results show some interesting and notable findings.

Undergraduate students appear to conceive of learning computer science in diverse ways with a greater emphasis on Programming and Calculating and practicing. These same students were least interested in the facets of Testing and Memorizing for learning computer science. It is unsurprising that Programming was a dominant factor in our study as we intentionally collected data from a range of computer programming courses at the institution. Also, the skill of Programming traces the computer science major from the early introduction courses (e.g., CS1) to the more advanced courses (e.g., compilers) that use programming as a skill to complete the learning activities. Analogous to prior research on student learning conceptions in computer science (Bruce et al., 2004; Hewner, 2012; Liang et al., 2015; Thuné & Eckerdal, 2009), programming plays a major role in how students conceive of learning within the field. Our results are also consistent with the prior study by Liang et al. (2015) that showed that students did not perceive of Testing and Memorizing as valuable for learning computer science. As noted by Kallia (2017), instructors use a wide range of assessment methods in computer science classrooms—not just traditional assessments that require rote memorization on a test.

When examining the ALCS, our data suggest that undergraduate students employ a Deep strategy to learning computer science involving strategies to make meaning of the subject and connect prior knowledge with new topics in computer science being learned. While the Deep strategy is rated as the highest factor, the Surface motive, in which students are motivated by extrinsic motivations such as getting higher grades or meeting teacher expectations appears to also drive student behavior more than the Deep motive in which students are motivated more by intrinsic factors like interest and curiosity. Notably, the lowest rated factor was the Surface strategy in which students remember rote information to pass examinations. These findings mirror those of Liang et al. (2015) with Taiwanese undergraduate students in computer science. Perhaps the ALCS span across international borders and cultures. However, to fully address that possibility, more data from across the globe must be collected and examined.

The findings from the Pearson correlations and stepwise multiple regression models among the factors on the COLCS and ALCS surveys suggest a clear pattern between conceptions of learning and ALCS. The Surface strategy was strongly related to Memorizing and Testing and negatively related to the remaining constructs. In contrast, the Deep strategy was strongly related to Calculating and practicing, Programming, Increasing one’s knowledge, Application and understanding, and Seeing in a new way and negatively related to Memorizing and Testing. After running the stepwise regression models, we can see that Testing (negative relationship) and Seeing in a new way (positive) were consistent for both the Deep motive and Deep strategy approaches. Memorizing was consistent in both the Surface motive and Surface strategy approaches with a positive relationship. These data show that low-level conceptions (e.g., Memorizing) and high-level conceptions (e.g., Seeing in a new way) can be generated based on these relationships. That is, conceptions of learning and ALCS are meaningfully related.

An important question to ask is what do these relationships mean to computer science education at the undergraduate level? Undergraduate computer science students with Deep motives and that employ a Deep strategy appear to have conceptions of learning unaligned with the test-centric focus of the computer science curriculum. Rather, these students appear to be more internally motivated by interest and curiosity and conceive of learning computer science as novel perspective for finding more reasonable explanations of the world related to computer science. Conversely, those students who have Surface motives and use Surface strategies are more externally motivated by passing the exam, meeting the degree requirement, getting a well-paid job, or pleasing family members. These students appear to employ weaker learning strategies like memorizing with a test-centric approach.

This study also sought to examine the individual differences of gender (male vs. female) and minority status (White vs. mon-White) across the factors on the COLCS and ALCS surveys. Employing a series of MANOVA models, only one construct showed a statistically significant difference: programming based on gender with male students having a slightly higher conception of learning than their female equivalents. Also notable, while not statistically significant, there was a potential relationship between White and non-White students on the use of the Surface strategy approach with non-Whites favoring the approach slightly more than Whites. As broadening participation among women and minority groups has been a stated objective of several computer science education grant programs by the National Science Foundation, understanding if women or minorities have different conceptions of learning and ALCS is an important research question. These preliminary findings suggest this research question may be worth expanding on a larger scale to fully understand whether these individual differences contribute to noticeable differences in learning conceptions and approaches.

Conclusion

What can we conclude about this research on conceptions of learning and ALCS? First, we have shown that there is variation in both these constructs in how college students both conceive and approach computer science. College students in this study appear to favor conceptions of Programming and Calculating and practicing over Memorizing and Testing. While extrinsically motivated, these college students appeared to favor a Deep strategy approach to learning computer science in which students use prior knowledge and deeper strategies to learn the topics. Second, there is a meaningful and palpable relationship between conceptions of learning and ALCS that manifests itself as both low-level conceptions (e.g., Memorizing) and high-level conceptions (e.g., Seeing in a new way) in relation to Surface strategy and Deep strategy, respectively. These relationships beckon further investigation in larger samples across institution types, national borders, cultures, and languages. Finally, there may exist some important differences between genders and minority status that deserve further attention from educational researchers and computer scientists. These differences may inform how we approach broadening participation in the field of computer science. Future research opportunities in this area are ripe with opportunity.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Karthikeyan Umapathy is an associate professor at the School of Computing, University of North Florida. He received his PhD in Information Sciences and Technology from the Pennsylvania State University. He is the 2018–2020 recipient of the FIS Distinguished Professor in Computing Award. He is a codirector of Florida Data Science for Social Good (FL-DSSG). He uses service learning as a means to help students understand that it takes more than technical expertise to be a good professional and excel in the job. He provides a real-world software development experience for students by identifying collaborative projects from nonprofit and small business organizations. His research interests are designing and developing complex information systems, analyzing web service standards, investigating innovative methods to teach computing concepts, and examining IT standardization processes.

Albert D. Ritzhaupt is an associate professor of Educational Technology and the associate director for Graduate Studies in the School of Teaching and Learning at the University of Florida. Dr. Ritzhaupt is an accomplished educational researcher and technologist. Dr. Ritzhaupt has published more than 100 journal articles, book chapters, technical reports, and conference proceedings and has presented his research at numerous state, national, and international conferences. Dr. Ritzhaupt currently serves as the editor for the Journal of Research on Technology in Education.

Zhen Xu is a PhD student of Educational Technology in the School of Teaching and Learning at the University of Florida. She is also pursuing a minor in Research and Evaluation Methodology with a quantitative focus. Her scholarship focuses on exploring, developing, and implementing frameworks, curriculum, and technologies to promote meaningful STEM learning in both K-12 and higher education settings and in formal, nonformal, and informal learning environments.

References

Bentler

P. M.

(1989) EQS structural equations program manual, Los Angeles, CA: BMDP Statistical Software.

Bentler

P. M.

Bonett

D. G.

(1980) Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin 88(3): 588.

Biggs, J. B. (1987). Student Approaches to Learning and Studying. Research Monograph. Australian Council for Educational Research Ltd., Radford House, Frederick St., Hawthorn 3122, Australia.

Bruce

Buckingham

Hynd

McMahon

Roggenkamp

Stoodley

(2004) Ways of experiencing the act of learning to program: A phenomenographic study of introductory programming students at university. Journal of Information Technology Education: Research 3: 145–160.

de Raadt, M., Hamilton, M., Lister, R. F., Tutty, J., Baker, B., Box, I., … Petre, M. (2005, July). Approaches to learning in computer programming students and their effect on success. In Proceedings of HERDSA 2005 International Conference: Higher Education in a Changing World, 28, 407–414.

Entwistle, N. J. (1998). Approaches to learning and forms of understanding. Teaching and learning in higher education, 72, 98.

Forte

Guzdial

(2005) Motivation and nonmajors in computer science: Identifying discrete audiences for introductory courses. IEEE Transactions on Education 48(2): 248–253.

Hewner, M. (2012). Student conceptions about the field of computer science (Doctoral dissertation). Georgia Institute of Technology, USA.

Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural equation modeling: a multidisciplinary journal, 6(1), 1–55.

10.

Kallia, M. (2017). Assessment in computer science courses: A literature review. Royal Society. Retrieved from: https://www.semanticscholar.org/paper/Assessment-in-Computer-Science-courses%3A-A-Review-Kallia/a82ff705b155b3e7fecd31e704e96d4108988ad9.

11.

Kember

Biggs

Leung

D. Y. P.

(2004) Examining the multidimensionality of approaches to learning through the development of a revised version of the learning process questionnaire. British Journal of Educational Psychology 74: 261–280.

12.

Lee

M.-H.

Johanson

R.-E.

Tsai

C.-C.

(2008) Exploring Taiwanese high school students’ conceptions of and approaches to learning science through a structural equation modeling analysis. Science Education 92: 191–220.

13.

Liang

J.-C.

Lee

M.-H.

Tsai

C.-C.

(2010) The relations between scientific epistemological beliefs and approaches to learning science among science-major undergraduates in Taiwan. Asia-Pacific Education Researcher 19: 43–59.

14.

Liang

J. C.

Y. C.

Tsai

C. C.

(2015) The assessment of Taiwanese college students’ conceptions of and approaches to learning computer science and their relationships. The Asia-Pacific Education Researcher 24(4): 557–567.

15.

Malakolunthu

Joshua

(2012) Learning approaches of undergraduate computer technology students: Strategies for improvement. Malaysian Journal of Learning and Instruction 9: 1–14.

16.

McCracken

(2001) A multi-national, multi-institutional study of assessment of programming skills of first-year CS students. SIGCSE Bulletin 33(4): 125–180.

17.

National Academies (2018) Assessing and responding to the growth of computer science undergraduate enrollments, Washington, DC: National Academies Press.

18.

Nunnally

J. C.

(1978) Psychometric theory, 2nd ed. New York, NY: McGraw-Hill.

19.

Obama White House. (2016). Computer science for all. Retrieved from https://obamawhitehouse.archives.gov/blog/2016/01/30/computer-science-all.

20.

Säljö

(1979) Learning in the learner’s perspective I: Some commonsense conceptions, Gothenburg, Sweden: Institute of Education, University of Gothenburg.

21.

Thuné

Eckerdal

(2009) Variation theory applied to students’ conceptions of computer programming. European Journal of Engineering Education 34(4): 339–347.

22.

Tsai, C. C., Ho, H. N. J., Liang, J. C., & Lin, H. M. (2011). Scientific epistemic beliefs, conceptions of learning science and self-efficacy of learning science among high school students. Learning and Instruction, 21(6), 757–769.

23.

Yan

Keusch

(2018) The impact of question and scale characteristics on scale direction effects. Survey Practice 11(2): 3126.

24.

van Rossum, E. J., & Schenk, S. M. (1984). The relationship between learning conception, study strategy and learning outcome. British Journal of Educational Psychology, 54(1), 73–83.

25.

Wan, T. (2017). Google, Facebook, Amazon among tech titans committing $300 million to K-12 computer science. EdSurge. Retrieved from https://www.edsurge.com/news/2017-09-26-google-facebook-amazon-among-tech-titans-committing-300-million-to-k-12-computer-science.

26.

Watson, C., & Li, F. W. (2014, June). Failure rates in introductory programming revisited. In Proceedings of the 2014 conference on Innovation & technology in computer science education (pp. 39–44). New York, NY: Association for Computing Machinery.

27.

White House. (2017). President Trump signs presidential memo to increase access to STEM and computer science education. Retrieved from https://www.whitehouse.gov/articles/president-trump-signs-presidential-memo-increase-access-stem-computer-science-education/.

28.

Wishart

(2005) A comparison of preferred learning styles, approaches and methods between information science and computing science undergraduates. Innovation in Teaching and Learning in Information and Computer Sciences 4(2): 1–21.