From EUP to EUSE What Weve Learned Margaret

From EUP to EUSE: What We’ve Learned Margaret Burnett Oregon State University

End-User Software Engineering: Beyond EUP • Today, end users can create programs. – e. g. , macros by demo, spreadsheets, email filtering rules, web authoring tools, graphical languages. . . • But what happens next? – Usually, bugs <many citations here>. – Compounded by overconfidence. • How to help solve this: learning from the end users. -2 -

End-User Software Engineering (cont) • What is EUSE? – Takes into account all of the software lifecycle. • Not just “create new programs”. – Requirements, design, programming, debugging, testing, maintenace, reuse, … -3 -

Interest in EUSE • Recent events – International media interest: NSF press release, ACM Tech. News, CNN. com, CBS. com, PC Magazine, IEEE Software editorial… – CHI’ 04/05/07/08 SIGs, ICSE’ 05 workshop, CHI’ 06 workshop, many papers at recent CHIs, ICSEs, and VL/HCCs, . . . • Here at ICSE’ 08 – This workshop, EUSE keynote! -4 -

What We’ve Learned about EUSE from End Users

Organization of this Topic • Spreadsheet debugging – and testing and code inspection – In light of gender differences • Unwitting end-user programmers • “Witting” end-user programmers • How tools can be more natural for end-user programmers -6 -

Spreadsheet Debugging • Bishop/Mc. Daid, Burnett et al. , Mc. Daid et al. • Via code inspection – Code inspection matters in debugging and spreadsheet debugging • to both males & females, but especially to females -7 -

Eight End-User Spreadsheet Debugging Strategies • Dataflow • Testing • Code Inspection • Specification Checking • Color Following • To-do Listing • Fixing Formulas • Spatial 8 -8 -

Code Inspection – Recommended by spreadsheet auditors. – Why females particularly like it: possibly information processing: • Comprehensive vs. heuristic processing. – In our studies, end-user programmers usually start with code insp, then move on to testing and other approaches if needed. -9 -

Code Inspection by EUPs – Business professionals outperformed business students at debugging. – Inspecting more cells led to more errors corrected. – Inspections focused more on top and left in groups, in bottom-line cells, and in formulas producing numeric outputs. - 10 -

What Else We Know about Male & Female Spreadsheet EUPs – For females: self-efficacy matters in spreadsheet auditing feature usage. • Not so for males. – Males tinker with new features more. • For males, this is sometimes good, sometimes not. • For females, tinkering is always good, but they don’t do it enough. - 11 -

What Else We Know about Spreadsheet EUPs (cont. ) – WYSIWYT Testing (Burnett/Rothermel) is achievable and useful for EUPs, especially males. – Mc. Daid et al. also showing some success with spreadsheet testing. - 12 -

Unwitting vs. Witting Software Development – Costabile et al. , Borggrafe et al. , Brandt et al. – A spectrum of end-user involvement in software. - 13 -

Unwitting Software Developers (cont. ) – Eg, “meta-design” (Fisher et al. , Costabile et al. , Borggrafe et al. ) • Designing/modeling, and/or working hand-in-hand with professional developers. – Eg, “Tailoring”: Some do not realize that they are doing software development. • Costabile et al. report this with medical domain experts • Petre/Blackwell report this with children game players. – Task-motivated (not programming-motivated), goaldirected, learning incrementally as they go. - 14 -

“Witting” EUPs • Segal’s study: – Scientists who explicitly program, but don’t view their software as an outcome or an asset. – Software not valued, not tested, not documented, not shared. • Clarke and Brandt et al. call EUP developers (and others) like this “opportunistic”. - 15 -

Opportunistic Programming • 5 characteristics (Brandt et al. ) – 1. Easier to build from scratch than to understand code libraries. • Happy to use tools new to them, if map well to their task. (Maybe more so with the males? ) – 2. When do reuse, use copy/paste. – 3. Process very iterative, debug/edit tightly interwoven. – 4. Code impermanent. – 5. Debugging challenges: less sophistication, multiple languages. - 16 -

Improving EUPs’ Abilities in Software Development • Teaching Software Engineering to EUPs (Umarji et al. ) – Bioinformatics developers don’t do software engineering (agrees with Segal’s findings). – They do share. They don’t like to “hand off” outside their domain. – Could benefit from education in QA, documentation, reuse. - 17 -

Related Theories and Support – Relates to Attention Investment (Blackwell) • Cost/benefit/risk. – Relates to Minimalist Learning Theory (Carroll/Rosson) • Paradox of “Active User”. – Strategies to support: • Domain-specific languages/closeness of mapping. • Surprise/Explain/Reward (Wilson/Burnett, …) – Connectts Curiosity to Minimalist Learning to Attention Investment. - 18 -

Helping Tools Be More “Natural” • “Natural” end-user software engineering (Myers et al. ) – Approaches/tools based on how end users think. • How do they debug? (led to Whyline) – Studies of people debugging: mostly they ask “why” and “why not”. – Help inform Whyline debugging tool: • for Alice (novices), for Word behavior (EUs), • and now for Java (bringing what learned about EUPs back to professional devs. - 19 -

Natural end-user software engr. (cont. ) • How Do UI Designers Design? – They iterate on appearance (no problem) and behaviors (big problem). • Complex, diverse behavior. – They build software prototypes to communicate design decisions. • Poorly supported by current tools. - 20 -

Natural end-user software engr. (cont. ) • Studies of professional developers too. – Can reveal insights into emerging issues that will affect EUPs soon too. – E. g. , studies of API usability, program comprehension. - 21 -

Conclusion: What we’ve learned • End-user programmers – Are not usually “baby” professional programmers, but do struggle with SE issues. – And have unique problems, different motivations, different langs/tools, gender diffs relevant to EUSE… • EUP studies provide critical info for tools. – And also can reveal new insights that open new avenues for supporting professional devs too. – And sometimes vice versa (apply with caution!) - 22 -

Question • What else do we know about end users that hasn’t come out yet? - 23 -