Difference between revisions of "Physics"

From LearnLab
Jump to: navigation, search
(Studies Conducted)
Line 57: Line 57:
 
*[[Self-explanation: Meta-cognitive vs. justification prompts|Self-explanation: Meta-cognitive vs. justification prompts (Hausmann, van de Sande, Gershman, & VanLehn, 2008)]]
 
*[[Self-explanation: Meta-cognitive vs. justification prompts|Self-explanation: Meta-cognitive vs. justification prompts (Hausmann, van de Sande, Gershman, & VanLehn, 2008)]]
  
*Nokes and Gadgil study at USNA, fall 2008.
+
*[[Analogical Scaffolding in Collaborative Learning|Analogical Scaffolding in Collaborative Learning (Gadgil &Nokes, 2008-2009)]]
  
 
In progress or planned:
 
In progress or planned:

Revision as of 17:28, 22 January 2009

Physics LearnLab Course

The Physics LearnLab Course (PLLC) is a research facility for studying how students learn introductory physics. It provides baseline data on student activities throughout the physics course, and it hosts specific research studies that measure the improvement in students’ learning caused by changes in the instruction. At this time, it is sited in the two-semester Introductory Physics courses at the US Naval Academy in Annapolis, MD and three courses at Watchung Hills Regional High School in Warren, NJ.

In order to increase the number of LearnLab sites, it is essential that we increase the number of students using Andes. During 2009, we plan to create a completely new web-based user interface. This will allow us to integrate Andes into WebAssign, the leading commercial provider of physics online homework making Andes easily accessible to over a hundred thousand students.

Students in PLLC classes use the Andes intelligent tutoring system to do their homework. Andes allows the PLLC to collect fine-grained data on student activity through the entire semester. The remainder of the course is taught the usual way, with lectures, labs, and a commercial paper-based textbook. In vivo experiments take place either by modifying Andes or by running studies during lab sessions that instructors have “donated” to the PLLC.

Studies Conducted

Summary of Studies
In Vivo Pull Out Lab Capacity
Course Run Planned Run Planned Run Planned Total # Sections Total # Students Max # Studies / Year Max # Students / Study
Physics 10 2 0 0 3 1 5 130 4 65

Capacity was determined by counting the number of students who solved more than 40 Andes problems in Fall 2007. There are about 25 students in a section and each LearnLab site has about 65 students.

Completed studies:

In progress or planned:

  • Comparing two homework systems, Sophie Gershman, 2008--2009.
  • Nokes and Gadgil lab study.

Achievements

From its inception in January 2005 to the present, we have achieved the following:

Content development milestones

  • The number of Andes problems assigned by instructors at the Naval Academy has increased from 58% to 100% in the Fall semester, and from 42% to 75% in the Spring semester.
  • We have increased the total number of working Andes problems from 350 to 556.
  • The number of physics principles has increased from 126 to 219. The number of rules in the physics “Knowledge Base” (the AI system) has increased from 619 to 915. The number of scalar quantities defined in Andes has increased from 85 to 126.
  • We shot videos of problems being solved—at least one per problem set—and revised many of the older videos. These act as worked examples. Students who view the videos in a problem set before solving any problems have a much easier time of it.

Enabling Technologies

  • We developed a way to run Andes under OLI. In particular, we found ways to get them to communicate through the USNA firewall, to upload log data and solution files, and to recover gracefully from most crashes.
  • We developed a method to control the data that the OLI gradebook exports to spreadsheets so that only the data that instructors wanted was exported in a format they specified.
  • Implemented “gating,” a method to force students to solve Andes problems in a pre-determined order. This was needed for the Sandy Katz experiment in fall 2006.
  • Andes raw logs can now be converted to the [DataShop] format at the knowledge component level (June 2007). The knowledge components associated with each correct student action (corresponding with a step) and must incorrect action (see transaction) is determined by Andes.

Log file analysis

The Andes log files represent a rich source of information about student problem solving but have not been studied in depth, outside the needs of specific experiments. We have begun to study the log files and begun to promote such work in the Physics Education Research (PER) community.

  • Studied time usage (how long does it take to apply a KC?) and time-on-task (are they really working?). Investigated whether time-on-task could be used as a metric for student learning of KC's.
  • Begun comparing the Log data to end-of-semester surveys administered at the USNA. The surveys were not anonymous, so individual survey results can be matched with the associated log files.
  • Conducted a workshop on log file analysis at PERC 2007. Two senior members of the PER community, Joe Redish and Gerd Kortemeyer, attended, expressed initial interest and corresponded with us after the conference, but no firm plans have been made.

Adoption of Andes

As of Fall 2008, Andes is being used at the following institutions:

  • St. Anselm college, Manchester NH (1 instructor).
  • US Naval Academy (1 instructor, several sections).
  • SUNY Fredonia (1 instructor).
  • Gannon University, Erie PA (1 instructor, several sections).
  • Conant High School, Hoffman Estates, IL (1 instructor).
  • Watchung Hills Regional High School, Warren NJ (2 instructors, several sections).

We see a shift in usage relative to previous years. Currently, the Naval Academy accounts for only 25% of our users; 32% of our users are now from High Schools.

We observe steadily growing use of Andes by individuals not enrolled in any OLI course. From January to April 2008, between 90 and 278 different users (some use is anonymous, precluding an exact count) solved a total of 1647 Andes problems. The previous semester, a total of 1260 problems were solved.

Advertising Andes in the physics community

We have focused our efforts on meetings of the American Association of Physics Teachers (AAPT) and the American Physical Society (APS) where we have presented numerous talks, posters, and a workshop.

These meetings generally do not publish proceedings.

More recently, we have begun promoting the Physics LearnLab at regional AAPT meetings:

  • Touring the Electromagnetic Spectrum (OSAPS 2008), Youngstown OH, March 2008. Vendor exhibit.
  • Central Pennsylvania Section of the American Association of Physics Teachers (CPS/AAPT), Lock Haven PA, April 2008. Vendor exhibit.
  • Fall meeting of the Arizona section of the AAPT, October 2008. Workshop for instructors.

In addition, we have presented Andes at other universities: Southern Methodist University (2006), the Ohio State University (2007), Rutgers University (2007), US Air Force Academy (2007), and the US Naval Academy (2007).

Publications on Andes

  • VanLehn, K., Lynch, C., Schulze, K., Shapiro, J.A., Shelby, R., Taylor, L., Treacy, D., Weinstein, A., and Wintersgill, M. The Andes Physics Tutoring System: Lessons Learned. International Journal of Artificial Intelligence and Education, 15 (3), 1-47.
  • Vanlehn, K., Lynch, C., Schulze, K., Shapiro, J. A., Shelby, R. H., Taylor, L., Treacy, D. J., Weinstein, A., and Wintersgill, M. C. The Andes physics tutoring system: Five years of evaluations. In G. McCalla, C. K. Looi, B. Bredeweg & J. Breuker (Eds.), Artificial Intelligence in Education. (pp. 678-685) Amsterdam, Netherlands: IOS Press.
  • Nwaigwe, A., Koedinger, K.,VanLehn, K., Hausmann, R. G. M. & Weinstein, A. (2007) Exploring alternative methods for error attribution in learning curves analyses in intelligent tutoring systems. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp 246-253. Amsterdam, Netherlands: IOS Press.
  • VanLehn, K., Koedinger, K., Skogsholm, A., Nwaigwe, A., Hausmann, R.G.M., Weinstein, A. & Billings, B. (2007). What’s in a step? Toward general, abstract representations of tutoring system log data. In C. Conati & K. McCoy (eds). Proceedings of User Modelling 2007.
  • VanLehn, K., & van de Sande, B. (in press) Expertise in elementary physics, and how to acquire it. In K. A. Ericsson (Ed.), Development of professional expertise: Toward measurement of expert performance and design of optimal learning environments.

Publications on PLLC experiments

  • Connelly, J. & Katz, S. (2006). Intelligent dialogue support for physics problem solving: Some preliminary mixed results. Technology, Instruction, Cognition, and Learning, 4, 1-29.
  • Ringenberg, M. & VanLehn, K. (2006). Scaffolding problem solving with annotated, worked-out examples to promote deep learning. In K. Ashley & M. Ikeda (Eds.), Intelligent Tutoring Systems: 8th International Conference, ITS2006. pp. 625-634. Amsterdam: IOS Press.
  • Chi, Min & VanLehn, K. (2007) The impact of explicit strategy instruction on problem-solving behaviors across intelligent tutoring systems. In D. McNamara & G. Trafton (Eds.) Proceedings of the 29th Annual Conference of the Cognitive Science Society. pp. 167-172 Mahwah, NJ: Erlbaum.
  • Chi, Min & VanLehn, K. (2007) Domain-specific and domain-independent interactive behaviors in Andes. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp. 548-550. Amsterdam, Netherlands: IOS Press.
  • Chi, Min & VanLehn, K. (2007) Porting an intelligent tutoring system across domains. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp. 551-553. Amsterdam, Netherlands: IOS Press.
  • Chi, Min & VanLehn, K. (2007) Accelerated future learning via explicit instruction of a problem solving strategy. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp. 409-416. Amsterdam, Netherlands: IOS Press.
  • Craig, S. D., VanLehn, K., Gadgil, S., & Chi, M. T. H. (2007). Learning from collaboratively observing videos during problem solving with Andes. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp. 554-556. Amsterdam, Netherlands: IOS Press.
  • Hausmann, R. G. M. & VanLehn, K. (2007). Explaining self-explaining: A contrast between content and generation. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp. 417-424. Amsterdam, Netherlands: IOS Press.
  • Hausmann, R. G. M. & VanLehn, K. (2007). Self-explaining in the classroom: Learning curve evidence In D. McNamara & G. Trafton (Eds.) Proceedings of the 29th Annual Conference of the Cognitive Science Society. pp 1067-1072 Mahwah, NJ: Erlbaum.
  • Katz, S., Connelly, J., & Wilson, C. (2007). Out of the lab and into the classroom: An evaluation of reflective dialogue in Andes. In R. Luckin, K. R. Koedinger & J. Greer (Eds.), Artificial Intelligence in Education 2007.
  • Nwaigwe, A., Koedinger, K.,VanLehn, K., Hausmann, R. G. M. & Weinstein, A. (2007) Exploring alternative methods for error attribution in learning curves analyses in intelligent tutoring systems. In R. Luckin, K. R. Koedinger & J. Greer (Eds.) Artificial Intelligence in Education. pp 246-253. Amsterdam, Netherlands: IOS Press.
  • VanLehn, K., Koedinger, K., Skogsholm, A., Nwaigwe, A., Hausmann, R.G.M., Weinstein, A. & Billings, B. (2007). What’s in a step? Toward general, abstract representations of tutoring system log data. In C. Conati & K. McCoy (eds). Proceedings of User Modelling 2007.
  • Hausmann, R. G. M., van de Sande, B., & VanLehn, K. (2008, May). Trialog: How Peer Collaboration Helps Remediate Errors in an ITS. Paper presented at the 21st meeting of the International FLAIRS Conference, Coconut Grove, FL.
  • Hausmann, R. G. M., van de Sande, B., & VanLehn, K. (2008, June). Shall we explain? Augmenting Learning from Intelligent Tutoring Systems and Peer Collaboration. Paper presented at the 9th meeting of the International Conference on Intelligent Tutoring Systems, Montréal, Canada.
  • Hausmann, R. G. M., van de Sande, B., van de Sande, C., & VanLehn, K. (2008, June). Productive Dialog During Collaborative Problem Solving. Paper presented at the 2008 International Conference for the Learning Sciences, Utrecht, Netherlands.

Current Status

The PLLC at the US Naval Academy is currently comprised of 3-5 sections (depending on the semester) of 25 students each. The sections are taught by Professors Mary Wintersgill and Ted McClanahan. At Watchung Hills Regional High School, the instructors are Sophie Gershmann and Brian Brown who teach three different levels of physics courses, mostly for Juniors and Seniors. The students use Open Learning Initiative (OLI) to access Andes, and the instructors use OLI to view gradebooks. Both high school and college students use Andes at home to do their regular homework assignments. Occasionally, Andes is used in class, but such “seat work” is not common.

Raw log data from Andes is stored on OLI servers. The raw data is periodically converted to DataShop format, but the conversion process is still not completely satisfactory, as some information is still available only from the raw log data. Researchers thus refer to both types of data.

All user identification is encrypted. The mapping between encrypted identities and student names is held by the Andes development programmer, Anders Weinstein. Instructors see only the students’ user identification before encryption; researchers see only the encrypted identities. Non-log data, such as hard-copies of midterm exams or audio files from verbal protocols, are collected as needed for specific experiments. They are anonymized by Anders Weinstein and stored in locked file cabinets or secure servers.

Although most experiments are in vivo experiments conducted in the PLLC courses, some studies are conventional lab studies. For instance, an experimenter might first run a study in the lab with paid volunteers and later do an improved version of the study in one or more PLLC classes.

Plans

Our major goal continues to be to expand the number of sites and instructors involved in the PLLC. There are simply not enough lab slots and students to meet the existing demand from PLLC experimenters. In order to increase involvement in the PLLC, we first need to increase the number of instructors using Andes in their courses, and make their experience a positive one.

Increase awareness of Andes

We need to increase awareness of Andes in the physics community. To date, we have focused our efforts on national meetings of the AAPT and APS. However, we plan to broaden our efforts:

  • We have begun to promote Andes at regional AAPT meetings and hope to expand this effort in the future.
  • We plan to arrange a summer school targeted mainly at regional high-school teachers of physics. Our long-term desire is for the summer school activity to eventually grow into a community of users consisting of both high school and college level instructors.
  • Continue visiting physics departments at other universities.
  • Publish PLLC-related research in the physics education journals.

Web-based delivery

Andes currently runs on Microsoft Windows machines as a Windows executable, requiring a software download/installation before it can be run. We lost at least two potential sites (Paul Perkins’ High School class in Bellevue WA and the US Air Force Academy) due to issues associated with this. In both cases, instructors were enthusiastic about Andes and assigned Andes to their students, but a significant number of students had troubles installing the software, getting it to run reliably, or did not have Microsoft Windows available to them. We believe we are losing many other potential clients due to this architecture. Thus, we have begun the development of a new web-based user interface to allow delivery of Andes as a true web application.

Improvements to Andes itself

Based on conversations with potential instructors as they view demonstrations of Andes and on instructors who have dropped Andes after using it, we have identified several aspects of Andes itself that we need to improve:

  • Instructors want a user interface that appears to be simple to learn. The new version of the user interface that we are developing will have a very simple design, making it similar to a generic drawing program (like Powerpoint).
  • Instructors want Andes to be a commercial product. In particular, they are worried about the long-term stability of the software product and that user support may be sporadic or unprofessional. The new user web-based interface will allow us to deliver Andes via our partners, WebAssign and LON-CAPA. Furthermore, we plan to offer Andes under an Open Source License, to ensure long-term availability and allow others to contribute to the future development of Andes.
  • Instructors want all reasonable student actions to be accepted. The new user interface will feature free text input, allowing greater flexibility.
  • Instructors want good, effective hints. We plan to make instructor evaluations of hint sequences an integral part of future workshops and summer schools. However, to really improve the hint quality would require that Andes maintain a model of the student across problems. This is one aspect of expert human tutoring that we can't capture with the existing system.
  • Other improvements requests that we hear regularly:
    • Allow sensitivity to lengths of vectors.
    • Allow vector equations (currently, Andes equations are all scalar).
    • Instructor control over policy for student actions that are correct but don't contribute to a solution.

Grading policy

Unfortunately, the current grading rubric is opaque and complicated and we are not always happy with the validity of the scores. There are two problems:

  • We don’t have any mechanism for an instructor to understand or modify the scoring rubric.
  • Some students become focused on raising their scores and, due to various weaknesses of (or incorrect inferences about) the scoring rubric, engage in behaviors that may raise their scores but do not constitute good problem-solving practice. For instance, a student will put in the final answer to a problem, and then go back and add problem-solving steps until their score is acceptably high.

Since one of the main goals of a grading policy is to encourage students to engage in productive problem solving behavior, any changes to the grading policy must be accompanied by log file analysis.

Supporting existing Andes users

There are a number of non-PLLC instructors using Andes in their classrooms as well as a number of users not affiliated with any OLI course.

  • Provide instructor support for setting up and running classes and user support for difficulties installing and running Andes.
  • Add instructor requested homework problems. We will continue our policy of adding new content based on instructor requests.
  • Add instructor requested problem types (such as graph drawing).
  • Fixing instructor reported bugs and complaints promptly. In particular, Andes sometimes gives hint sequences that are not helpful. Also, it sometimes won't accept solution steps that instructors would allow.
  • Develop log file analysis to detect ineffective hint sequences, common student difficulties, and plain old bugs.
  • Eventually, hold some instructor workshops for existing instructors, so that they feel part of the Andes development process and connect with other Andes users.

Log file analysis

The knowledge components (KCs) used by Andes generally do not produce the nice learning curves that one would expect, which makes it problematic for experimenters to use them as dependent measures. We suspect that the present physics KCs implicitly contain the knowledge needed for applying a principle within a problem context along with the principle itself. Thus, when a KC that has been practiced several times in simple problems is used for the first time in a complex problem, the associated assistance score may be higher than expected. In fact, it is common practice in physics homework assignments to exercise students in applying physics principles in widely varying problem contexts. Thus, as the problem context varies, the difficulty of applying our present KCs vary widely, resulting in widely varying assistance scores. We have been doing data mining to test this hypothesis, but this has been a backburner activity and is moving slowly.

Here are some continuing activities associated with log files:

  • Download log files from OLI, anonymize them, and load them into the DataShop.
  • Andes raw logs can be converted to the DataShop format, but the converted logs often do not have the right information in them for the kinds of analysis experimenters want to do, so the converter scripts must be changed.
  • Finish investigating whether time spent can be a useful metric of student learning.
  • Continue investigating why the present KC’s produce learning curves that do not match current theoretical predictions.
  • Continue promoting Log file analysis as an interesting area of research, especially for those interested in developing cognitive models of student learning.