PaperlayerView exact revision on arXiv

Source-linked AI summary

Remixing as a Pathway to Computational Thinking

Sayamindu Dasgupta, William Hale, Andrés Monroy-Hernández, Benjamin Mako Hill

arXiv:1605.08766v1cs.CYcs.HCcs.SI

TL;DR

The paper tests whether remixing is associated with learning in Scratch, using project, user, and code data. It finds positive associations between remixing and programming repertoires, and between exposure to computational-thinking concepts through remixing and later use of those concepts.

  • Problem

    Although remixing was theorized as a pathway to learning, the authors knew of no empirical test of that theory in remixing communities.

    Source: BACKGROUND · p. 2

  • Method

    The study combines Scratch user, project, and interaction metadata with algorithmic analyses of project files.

    Source: DATA AND MEASURES · p. 3

  • Results

    Programming-block repertoires were positively associated with remixing when controlling for published code, and use of six computational-thinking concepts was positively associated with exposure through remixing.

    Source: DISCUSSION · p. 9

  • Takeaways & Limitations

    The findings support the theory that users can learn through remixing and support remixing as a pathway to learning.

    Source: DISCUSSION · p. 9

    Without exogenous variation in remixing, the study cannot establish that remixing causes learning, and omitted confounders may remain.

    Source: LIMITATIONS · p. 8

Abstract

from arXiv · show

Theorists and advocates of "remixing" have suggested that appropriation can act as a pathway for learning. We test this theory quantitatively using data from more than 2.4 million multimedia programming projects shared by more than 1 million users in the Scratch online community. First, we show that users who remix more often have larger repertoires of programming commands even after controlling for the numbers of projects and amount of code shared. Second, we show that exposure to computational thinking concepts through remixing is associated with increased likelihood of using those concepts. Our results support theories that young people learn through remixing, and have important implications for designers of social computing systems.

INTRODUCTION

Scratch was designed to help young people learn programming by downloading, studying, and modifying others’ projects. This paper tests that theory with a large longitudinal dataset and reports results that broadly support learning through remixing.

INTRODUCTION

  • Scratch was designed as a platform for young people to learn programming through remixing, with shared projects licensed for unrestricted reuse and modification.Novices could download, study, and modify projects made by more skilled users.Source: INTRODUCTION · p. 1
  • The theory rests on the idea that engaging with others’ work exposes users to different skills, knowledge, and experiences that may promote learning.The paper situates this idea in programming and computational-thinking education.Source: INTRODUCTION · p. 1
  • The study tests whether remixing is associated with increased programming skills and computational thinking learning, using 2,426,894 projects created by 1,068,502 users.The authors report broad support for the idea that young people learn through remixing.Source: INTRODUCTION · pp. 1–2

BACKGROUND

Remixing has been proposed as a route to participation, production, and learning, but its learning effects had not been empirically tested in remixing communities. The paper frames two tests around programming-repertoire growth and concept-specific computational-thinking use.

BACKGROUND

  • Remixing is described as reworking and combining existing creative artifacts, and its advocates link it to participation, aggregated production, and learning.Other commentators have criticized remixes as derivative, uninteresting, or poor in quality.Source: BACKGROUND · p. 2
  • Prior empirical research examined remixing communities, authorship, copyright, and output quality, while the authors knew of no empirical test of remixing as a pathway to learning.Earlier Scratch learning studies were mostly qualitative case studies, and measuring learning in informal settings is challenging.Source: BACKGROUND · p. 2
  • Constructionism motivates Scratch’s social learning rationale, while computational-thinking concepts include basic building blocks such as sequences, loops, and conditionals.Constructionist theory emphasizes designing and constructing public entities, with learning made more meaningful and effective in a social context.Source: BACKGROUND · p. 2
  • The paper tests whether users’ programming repertoires grow more when they engage in more remixing activity.Repertoire growth is modeled as accumulation of programming tokens over time.Source: BACKGROUND · pp. 2–3
  • The second hypothesis predicts that remixing projects using a particular computational-thinking concept is associated with greater likelihood of first using it in a de novo project.The authors test this separately for each of Brennan and Resnick’s computational-thinking concepts.Source: BACKGROUND · p. 3

EMPIRICAL SETTING

The study uses Scratch, a public community where users create, share, and remix interactive media with a visual, block-based programming environment. Its analysis covers the community before browser-based editing was introduced.

EMPIRICAL SETTING

  • Scratch was a large public community for creating, sharing, and remixing interactive media, with more than 7 million users, 10 million projects, and 2.6 million remixes.It was described as the largest community dedicated to remixing and the largest informal programming-learning community for young people.Source: EMPIRICAL SETTING · p. 3
  • Scratch programs are built by dragging and dropping visual blocks, which can be combined into scripts to control on-screen sprites.The environment includes more than 120 distinct blocks.Source: EMPIRICAL SETTING · p. 3
  • The analysis covers Scratch activity before the web-based editing environment was introduced; most users self-reported ages 8–17, with a median age of 12 for new accounts.The age information and analysis window describe the setting, not necessarily every user or project in the dataset.Source: EMPIRICAL SETTING · p. 3

DATA AND MEASURES

The study combines Scratch project and user metadata with analyses of project code, then measures programming-block repertoire and use of specific computational-thinking concepts. Analyses focus on users with at least two projects and account for project activity, code volume, downloads, and other user characteristics.

DATA AND MEASURES

  • 2,426,894 projects shared from 2007 through 2012 form the dataset, combining Scratch metadata with algorithmic analyses of project files.Projects are the unit of analysis.Source: DATA AND MEASURES · p. 3
  • 173,053 users with at least two projects make up the within-user analysis sample.The study also omits projects with missing, corrupt, or technically unavailable data.Source: DATA AND MEASURES · pp. 3–4
  • Programming repertoire is the number of block types ever used, while concept measures record whether a project demonstrates a computational-thinking concept.The study maps Scratch blocks to concepts; parallelism is detected when multiple event-handling blocks listen for the same event, and sequences are excluded because nearly every project contains them.Source: DATA AND MEASURES · p. 4
  • Models account for project counts, comments, age, account experience, and code volume, including controls intended to address code-light coloring-contest remixes.The study also uses project downloads as a control.Source: DATA AND MEASURES · pp. 4–5
  • Downloaded projects may expose users to code without producing a publicly shared remix, so the study’s remix measure is conservative relative to broader definitions.The authors urge interpretation of results in this context.Source: DATA AND MEASURES · p. 5
  • Because participation measures are highly skewed, the study log-transforms independent variables to make their distributions more nearly normal.Table 2 reports user-level medians, means, standard deviations, and ranges.Source: Table 2 · p. 4, DATA AND MEASURES · p. 5

ANALYSIS

The analysis uses within-user panel regressions to test whether remixing is associated with block-repertoire growth, and Cox models to examine first use of individual concepts. Time in the survival models is measured by projects shared, and plots show the share of users who have not yet demonstrated each concept.

ANALYSIS

  • H1 uses user-fixed-effects panel regressions, with an observation for each de novo project shared by users with at least two projects.The models include user age, experience, total blocks, and downloads as controls.Source: ANALYSIS · p. 5, Table 3 · p. 6
  • The exploratory H1 models add controls in sequence: M1 omits total blocks and downloads, M2 adds total blocks, and M3 adds downloads.The sequence tests whether code-light coloring-contest activity affects the remixing estimate.Source: ANALYSIS · p. 5
  • H2 uses Cox proportional-hazards models, with each observation representing a spell between projects and project count as the time variable.Separate models examine conditionals, data, events, loops, operators, and parallelism; concept-specific remix counts are included alongside total remixing and other controls.Source: ANALYSIS · pp. 5–6
  • Figure 3 plots the share of users who have never used a concept against the number of de novo projects, with users censored when they stop contributing before using it.The curves provide a non-parametric view of concept uptake over project count.Source: ANALYSIS · pp. 5–6

RESULTS

Users who remix more have larger programming-block repertoires after accounting for code volume, while remixing projects with computational thinking concepts is associated with using those concepts in later projects.

RESULTS

  • A log-unit increase in shared remixes corresponds to 1.4 additional repertoire blocks in M2, or 0.25 in M3 after controlling for downloads.For a prototypical M2 user with 100 de novo projects, the predicted repertoire is 81 blocks with no remixes and 86 with one remix per three de novo projects; the M3 difference is about one block, more than 1% of repertoire.Source: RESULTS · p. 6
  • Remixing one log-unit more projects with conditionals corresponds to 1.34 times the risk of first using conditionals in a de novo project, and remix effects are positive across all six CT concepts.Source: RESULTS · pp. 6–7, Table 4 · p. 7
  • Model estimates show higher use-risk for users who shared three concept-matching remixes than for users who shared none, with other variables held at their sample medians.Source: RESULTS · p. 7, Figure 4 · p. 8
  • The proportional-hazards assumption may be violated, so the authors caution readers to remain skeptical about specific model parameter estimates.Residual plots were largely flat, but sometimes decreased toward zero as more projects were shared.Source: RESULTS · p. 7
  • Downloads are positively associated with repertoire across models, while comments have mixed associations and raw remix counts are negative for concept use after controlling for concept-specific remixes.The authors suggest that social remixing such as coloring contests may help explain the negative association.Source: RESULTS · p. 7
  • More remixing is associated with larger block repertoires after accounting for code volume, and remixing projects with particular CT concepts is associated with later use of those concepts.Source: RESULTS · pp. 7–8

LIMITATIONS

The study’s learning measures may reflect confounds or copied code rather than understanding, and behavioral measures cannot reveal what users learned internally. Its findings also have uncertain generalizability beyond Scratch, especially given that most users share few projects.

LIMITATIONS

  • The observational design cannot establish that remixing causes learning, and omitted controls may leave correlated factors unaccounted for.The authors control for popularity, experience, age, and technical sophistication, but lack exogenous variation in remixing.Source: LIMITATIONS · p. 8
  • Users may copy blocks without understanding them, but excluding concepts found in scripts identical to prior remixes leaves positive, significant results in all six cases.This alternative threshold discarded 2–15% of projects previously associated with a concept and moderately reduced parameter estimates.Source: LIMITATIONS · p. 8
  • Different initial skills and repertoires may affect both remixing and growth, although controlling for initial block repertoire did not substantively change results.Source: LIMITATIONS · p. 8
  • Scratch’s multiple sprites and green-flag event handlers can make parallelism appear unintentionally; restricting measurement to parallel scripts within one sprite strengthened the estimate.Source: LIMITATIONS · pp. 8–9
  • Behavioral measures cannot reveal what users learned internally, and other learning metrics could produce different results or conclusions.Source: LIMITATIONS · p. 9
  • The findings may not generalize beyond Scratch, where most users share only a small number of projects and may engage briefly.The authors note that short engagement may constrain the study’s methodological and substantive implications.Source: LIMITATIONS · p. 9

DISCUSSION

The authors find positive associations between remixing and programming-block repertoires, and between exposure to six computational-thinking concepts through remixing and users’ demonstration of those concepts. They argue that remixing can support learning, while noting modest effects and difficulty promoting remixing in informal settings.

DISCUSSION

  • Users’ programming-block repertoires are positively associated with remixing, and exposure to six computational-thinking concepts through remixing is positively associated with demonstrating them.The repertoire association holds controlling for the total amount of code published; the authors report robustness to potential validity threats.Source: DISCUSSION · p. 9
  • Estimated effects are relatively small, while Scratch’s unconstrained remixing is dominated by genres such as coloring contests that designers did not anticipate.The authors suggest stronger effects might be possible in more structured contexts.Source: DISCUSSION · p. 9
  • Teachers could structure remixing around code-intensive projects or particular concepts, and informal platforms could direct activity toward selected source material.The authors cite formal learning combined with Scratch, administrator direction on other platforms, and Scratch collaboration camps as examples.Source: DISCUSSION · p. 9
  • The study tests a widespread theory and proposes two quantitative methods for measuring computational thinking that may inform future Scratch research and other informal learning environments.Source: DISCUSSION · p. 9
  • Remixing may be more effective for engaging users with loops than with operators and data, suggesting opportunities to foster engagement with a wider range of concepts.The authors offer project popularity and differences in concept learnability within Scratch as possible explanations.Source: DISCUSSION · p. 9
  • Promoting remixing requires prolonged engagement and remains difficult in informal settings, with trade-offs that its association with learning does not resolve.The authors describe the study as a first step and call for further critique and development of its theories, measures, and approach.Source: DISCUSSION · pp. 9–10
Loading 1605.08766v1…