Arduinos and Games: Two Days of Computer Science for Teachers

The first day opened with a stack of about 50 sticky notes, each naming a city. The teachers had to sort them into regions of the country, with one person taking notes in, one person handing them out, and no talking once the notes started moving. After several tries they arranged themselves in a diamond and finished in under a minute. Only then did they learn what they had built: an algorithm, a set of steps that a computer could follow. Press Play on each formation above to see why the diamond wins.

The model gives each step a time: one second to pass a note, two seconds to read a city and decide its region, one second to drop it on a pile. One sorter needs 48 × 3 = 144 seconds. A line of five people is slower still, because the last person does all the reading. The diamond splits the reading among four people, so the job takes about as long as dealing the notes. The paper does not report these timings; they are the page’s, chosen to show the idea.

The activity is a parallel algorithm taught without a computer: the diamond is a scatter (input deals work), parallel map (four sorters classify), and gather (output places). The model makes its bottleneck visible; once reading is parallel, throughput is set by the slowest serial stage, the input person dealing one note per second, so the diamond finishes in 52 model seconds. The teachers did not know beforehand that the game simulated an algorithm, and the debrief connected it to efficiency, optimization, and what code lets astronomers and other STEM specialists do. The paper reports the formation and the time (under a minute), not a step model; this timeline is the page’s illustration.

Each day mixed playing with talking. Teachers moved between “try-it” stations at their own pace while experts in computer engineering, astronomy, and education asked how each activity could work in their classrooms. In March, six of the 11 teachers said they planned to use the NetLogo simulations, and five named the board games or the sorting game as new ideas they would take back.

NetLogo lets you build a world of small “agents” (wolves and sheep, burning trees, gas molecules, or planets) and watch what happens when each follows simple rules. An Arduino is a small programmable board that reads sensors and controls lights and motors. A Makey Makey turns anything that conducts electricity, even a piece of fruit, into a computer key. Every station in the list connects coding to something a science class already studies.

The March counts come from open responses rather than a closed survey, so they show what teachers volunteered: six of 11 named NetLogo, five named Robot Turtles or the sorting game, and six answered the improvement question with “Nothing” or a compliment. Suggestions were individual (fewer activities, a computer science expert on hand for coding questions, time to review SparkFun resources, mobile app development), and varied from teacher to teacher.

The June teachers rated the day 4.4 out of five. All of them said they were engaged more than three quarters of the time, all found the hands-on activities meaningful, and all planned to use at least some of the material with their own students. In their own words, the activities helped students “to visually see and experience” coding and showed “how a computer works.”

The paper reports these results as summary statements rather than item tables: a mean rating of 4.4, engagement above 75 percent of the time for all teachers, at least 70 percent “very useful” for every activity, and unanimous ratings of meaningful and intent to use. The abstract attributes the same figures to all 33 participants, while the results section attributes them to the 22 June teachers; the page follows the results section. Teachers also rated the collaborative lesson-planning time “extremely meaningful and productive.”

Twenty of the June teachers took the same short computer science quiz before and after the day. Their average went from 2.45 to 8.70. The quiz’s top score is not stated in the paper, but most teachers ended near the top and almost all started near the bottom.

The whiskers show the standard deviation, a measure of how spread out the scores are. Before the day, scores were spread widely (some teachers knew a little, most knew almost none); afterward they were tightly packed near the top. A gain of 6.25 points on a spread that small is far larger than chance could produce: the paper’s test gives p below 0.0001.

A paired t-test compares each teacher with themself, so it tests the average of the 20 individual gains against zero. The printed correlation between pre and post scores is low (0.14): knowing where a teacher started says little about where they finished. Move r to see how t depends on it. With the printed values, the page computes t = 13.67; Table 3 prints 13.64, and the rebuilding below explains the difference as rounding.

The paper names “effect size” without a formula. Of the common conventions, only the repeated-measures d (d_rm = t√(2(1 − r)/n), from Dunlap and colleagues and from Morris and DeShon) returns 4.008; d_z (gain over the standard deviation of gains) gives about 3.05 and d_av (gain over the average standard deviation) gives about 4.08. Every convention describes a gain several standard deviations wide. The identification is the page’s and is listed for the authors’ review.

Rebuilding Table 3

If each teacher’s score is a whole number, Table 3 fixes the sums completely. The mean of 2.45 for 20 teachers means a total of 49, and only one sum of squares (193) gives a standard deviation that rounds to 1.96; likewise 174 and 1530 for the post-test. Among the sums of products that the score lists allow, only 431 gives both r = 0.14 and t = 13.64. Every class consistent with the table therefore has the same statistics to full precision: t(19) = 13.643, p = 2.9 × 10⁻¹¹, r = 0.137, and d_rm = 4.0085. The individual scores remain unknown; the demo draws one consistent class at a time.

• Teachers learned computer science basics quickly when the activities were hands-on and tied to their subjects: average quiz scores rose from 2.45 to 8.70 in one day.
• The June teachers rated the day 4.4 out of five, and all of them planned to use some of it.
• In March, more than half planned to use NetLogo simulations in class.
• Teachers asked for time to practice, try, and fail, just as their students would.
• Astronomy gave teachers a reason to try coding.

• Large immediate content gains. Pre-test 2.45 (SD 1.96), post-test 8.70 (SD 0.92), N = 20, r = 0.14, paired t = 13.64, p = .0000, effect size 4.008 (Table 3, June 29).
• High ratings in June. Session mean 4.4 of 5; all 22 engaged more than 75 percent of the time; at least 70 percent rated each activity very useful; all rated the hands-on activities meaningful and intended to use some of the content.
• Uptake in March. Six of 11 planned to use NetLogo simulations; five named Robot Turtles and the “What is coding?” activity as new ideas.
• Design lessons. Teachers need safe time to try tasks, fail, and retry; collaboration with peers and experts supports learning; astronomy serves as an entry point to computer science and engineering.
• Next steps named by the authors. Measure K-12 student learning and interview teachers after the PD ends.

Most K-12 teachers never studied computer science, yet schools increasingly expect them to bring it into science and math classes. This study shows a way in: start with games and objects people can touch, connect them to a subject the teacher already loves (here, astronomy), and give teachers time to try things before asking them to teach them.

The paper argues that computer science and engineering can live inside existing K-12 science and math lessons when PD models the same problem-based, hands-on approach teachers are asked to use. The unplugged sorting game is the clearest example: it teaches algorithms, efficiency, and optimization with sticky notes, and it travels to any classroom for the price of a pad of sticky notes. Ultimately, the recommendation is structural: build safe practice time and expert access into the PD, not only content.

The evaluation documents a low-cost design (unplugged activities, board games, and inexpensive kits) that teachers rated highly and planned to adopt, with an immediate content gain that chance cannot explain (p < 0.0001). The single-group design, a same-day post-test, and the surrounding 23 PD days limit causal claims about the two days themselves, which the authors acknowledge; the case for the approach rests as much on the uniform intent to use as on the test.

Notes from building this page

• Effect-size convention. The printed 4.008 matches d_rm = t√(2(1 − r)/n): 4.0085 from the recovered sums, and 4.0077 from the printed t with the recovered r. Naming the convention matters here, because d_z (3.05) is the natural choice for a paired test and sits almost a full unit lower.
• Whole-number scores. The reconstruction assumes each teacher’s score is a whole number. Under that assumption the five sums are unique, so the table reproduces exactly. The maximum score is not printed; with a maximum of nine, two post-test lists fit (one has 18 teachers at nine and two at six), and with a maximum of ten, ten lists fit (for example, 12 at eight, two at nine, and six at ten). Either way most teachers sit at or near the top on the post-test, so the test’s ceiling, not the teachers, limits the post-test spread.
• Low correlation. With r = 0.14, pre-test scores barely predict post-test scores. A near-ceiling post-test produces exactly this pattern, and it also means the paired design adds little power over an unpaired comparison here.
• Same-day post-test. Table 3 labels both tests “June 29,” which reads as a post-test at the end of the day. The gain then measures immediate recall of material taught that day; the authors list the lack of follow-up as a limitation.
• Which participants. The abstract attributes the 4.4 rating, 75 percent engagement, and 70 percent “very useful” to all 33 participants; the results section attributes them to the 22 June teachers. Whether any teacher attended both days, so that 33 counts attendances rather than people, is not stated.
• Program size. This paper counts 41 LASSI teachers across 2014 and 2015; the 2021 Education Sciences paper on the same program (Borowczak and colleagues, “I Need Time for Practice”) counts 30. The difference may reflect who completed the full program.
• Small details. The paper spells Khan Academy as “Kahn Academy” and calls Codecademy “Code Academy”; the content questionnaire appears only as Figure 1’s title, and Table 1’s integrated-STEM example is not printed in the text; p is printed as “.0000” rather than as a bound.

Try this in class

1. Write the names of about 50 cities on sticky notes. Choose one student to take notes in and one to hand them out, and tell the class to sort the notes into four regions without talking once the notes move. Time each try.
2. Let the class change its formation between tries, then compare the times with the “What is coding?” demo: which person is the bottleneck in each formation?
3. Write the class’s best formation as numbered steps, the way a computer program would follow them, and discuss what would change with 500 notes instead of 50.