160 likes | 248 Views
From chaos to goals: Requirements change can be predicted. Johan F. Hoorn Vrije Universiteit Computer Science Information Management and Software Engineering jfhoorn@cs.vu.nl. Contents. Status Problem Analysis Model Method Case Results Conclusions. Johan F. Hoorn, 2004.
E N D
From chaos to goals: Requirements change can be predicted Johan F. Hoorn Vrije Universiteit Computer Science Information Management and Software Engineering jfhoorn@cs.vu.nl
Contents • Status • Problem • Analysis • Model • Method • Case • Results • Conclusions Johan F. Hoorn, 2004
Status • Sept. 1, 2001 – Aug. 31, 2005 • Supervisors: Gerrit van der Veer Hans van Vliet • Four international publications • Industries: • Human-Machine Interaction Johan F. Hoorn, 2004
Business model 1 Traditional office Mainframe with thin clients Business model 2 Flexible workplace Laptops with bluetooth A change request during development is extremely costly and frustrating Problem • Requirements change Johan F. Hoorn, 2004
From chaos to goals: Nobody knows how requirements change can be predicted
Analysis • Where do change requests come from? Business model 1 Business model 2 • Change in business sub goals - Main goals: Profit - Sub goals: Cost-effectiveness, efficiency • How come business goals change? • Change in sub goals (strategic management) - Main goals: Earn my living - Sub goals: Fire employees (not me), improve IT to guarantee same output Johan F. Hoorn, 2004
Model • Change of Stakeholder Requirements (CoStaR) (Hoorn & Van der Veer, 2003a; 2003b) One of the hypotheses: Valence Requirements Goals Stakeholder evaluation: Does a system feature support my goals? Does a system feature obstruct my goals? (after Frijda, 1986) Johan F. Hoorn, 2004
Method • REquest, the Requirements Engineering questionnaire General approach: Items that combine - a must or a won’t requirement, with - support or obstruction of - a goal to achieve with the system or a goal state to avoid, scored for agreement on a 6-point Likert scale Johan F. Hoorn, 2004
Case • Eighteen managers of a logistic warehouse management system must/won’t support/obstruct goal approach/avoid E-mail ordering increases efficiency E-mail ordering decreases efficiency E-mail ordering increases inefficiency E-mail ordering decreases inefficiency Paper ordering forms increase efficiency Paper ordering forms decrease efficiency Paper ordering forms increase inefficiency Paper ordering forms decrease inefficiency Example items Johan F. Hoorn, 2004
Results (1) • Original hypothesis: Valence Requirements Goals - Indeed, goals, valence, and requirements all evoked significant effects on agreement to requirements statements Johan F. Hoorn, 2004
Results (2) ↑ Grand mean agreement 5 4 3.67 (1.14) 2.78 3 2.5 2.41 (1.04) 2.19 (.96) 1.8 (.98) 2 (1.44) (1.09) 1 0 Goals (to Goals (to Requirements Requirements Valence Valence approach) avoid) (must have) (won't have) (support) (obstruct) MANOVA (must vs won’t) * (support vs obstruct) * (goal approach vs avoid) Pillai’s Trace = .51, F(2,16)= 8.40, p= .003, ηp2= .51 Johan F. Hoorn, 2004
- Goal-driven RE models should be unipolar Goals (approach) Requirements (must have) Valence (support) Valence (obstruct) Goals (avoid) Requirements (won’t have) Results (3) • Original hypothesis: Valence Requirements Goals • Bipolar conception does not hold • Regression: R2= .03, R2adj= -.03, F(1,16)= .47, p= .504 Johan F. Hoorn, 2004
Results (4) • However, original structure should be completely revised Goals Requirements (to approach) (won’t have) Valence (support) Valence (obstruct) Requirements Goals (must have) (to avoid) Johan F. Hoorn, 2004
no predictive power Yet, valence does have influence. Requirements vs. goals: Parameter coefficient= -.56, t= -4.04, p= .001, ηp2= .49 no predictive power 70%!! 90%!! Results (5) R2= .79, R2adj= .70 F(5,12)= 9.01, p= .001 Goals (approach) Requirements (won’t have) Valence (support) Valence is a moderator! Valence (support) Goals (avoid) Requirements (must have) R2= .93, R2adj= .90 F(5,12)= 30.30, p= .000 Johan F. Hoorn, 2004
Conclusions (1) • RE should be oriented to goals • Requirements validation should be done • with structured questionnaires • (e.g., REquest) • Goals to achieve predict won’t requirements • Goal states to avoid predict must requirems • Like the weather, valence does not predict • mood (i.e. agreement) but it does influence it Johan F. Hoorn, 2004
Conclusions (2) • Most important RE questions are: • What are the things you want to achieve with the system? • What should the system NOT have to support that? • What are the things you want to avoid • with the system? • What should the system offer to support that? Johan F. Hoorn, 2004