Environment and Planning B: Planning and Design, 1993, volume 20, pages 333-345 Fixation effects: do tey exist in design problem solving? A T Purcell, P Williams, J S Gero, B Colbron Department of Arcitectural and Design Science and Key Centre for Design Quality, Sydney University, Sydney, NSW 2006, Australia Received 1 Marc 1992 Abstract. Designing involves te use of a number of different types of knowledge wic vary from abstract knowledge to knowledge about pysical forms and teir attributes. Previous researc ad demonstrated tat pictorial representations of an example design presented as part of te statement of a design problem resulted in designs tat exibited te caracteristics of te pictorial example. Tis effect was referred to as design fixation. However, subsequent attempts to replicate te effect were largely unsuccessful wit wat appeared to be fixation under some conditions being closely related to familiarity wit existing examples of te design problem. Tere were, owever, oter differences between te two experiments, particularly in terms of te level of experience of te designers and te design discipline of te participants. In te experiment reported, familiarity wit existing examples of solutions of a design problem was controlled by coosing an appropriate problem from te set of problems used in te original researc and designers of te same discipline and level of experience were used togeter wit designers from different disciplines. In addition to te pictorial representation of a design solution used in te original researc, an additional pictorial example was included. Te results demonstrated clear fixation effects wen bot te discipline of te designers and te pictorial example were te same as tose used in te original experiment. However, no effect of te pictorial example occurred wit designers of different disciplines or te oter pictorial example for designers from any discipline. Possible reasons for tese results are discussed, wit particular reference to te effects of a matc between te principles used in te design solution and te discipline of te designers. Introduction Designing an object involves te use of a number of different types of knowledge. Typically we tink of knowledge as consisting of facts and principles tat can be expressed, for example, in linguistic or symbolic form. However, objects are made up of pysical elements consisting of particular materials wic ave properties, suc as sape, wit te elements arranged in particular ways to form te complete object. Tese caracteristics of objects constitute anoter possible form of knowledge (perceptual rater tan abstract) tat can be used in te design process. Te relevance of tis type of knowledge to design is apparent if te relationsip between te statement of a design problem and te outcome of te design process is considered. Generally te description of wat is to be designed is expressed in linguistic or symbolic terms. A consequence of tis is tat te response to te problem can be satisfied by a variety of possible pysical forms. Tis is because te facts and principles used to formulate te description do not uniquely map into sets of materials and te way tey are put togeter and, as a result, do not specify a single pysical form. However, te designer must arrive at a particular form wic meets te requirements of te design description. As a consequence, designing involves moving from knowledge about facts and principles wic relate to te description of a design problem to pysical forms wic express particular facts and principles wic are relevant to te problem. Te designer uses knowledge about facts and principles in order to understand te design problem, combined wit knowledge about pysical forms and teir properties and te ways in wic
334 A T Purcell, P Williams, J S Gero, B Colbron tey can be aggregated in order to move from te description of te problem to potential pysical solutions. Te experiment to be reported is concerned wit te relationsip between tese two types of knowledge in te process of design. Jansson and Smit (1991) refer to tese types as conceptual and configurational knowledge. Tey investigated te relationsip between te two by asking different groups of advanced undergraduate or practising mecanical engineers to solve a number of design problems under one of two conditions. Te control group were given a typical statement of te problem in verbal form, wereas te oter group were given te statement of te problem togeter wit a drawing of a potential solution to te problem. Te drawing was presented in te guise of a guide to te format in wic tey were to present teir designs. Te control group obtained conceptual knowledge about wat was to be designed and te oter obtained tis conceptual knowledge togeter wit configurational knowledge of a possible solution. Jansson and Smit found tat te groups sown pictures of possible design solutions produced more designs wit te caracteristics of te pictured examples for all of te problem types attempted tan te control group. In a number of te design problems, te pictured example deliberately included design features wic were inappropriate given te verbal statement of te problem and even tese features of te designs were produced more frequently tan in te control group. Tey refer to tis effect as design fixation, to igligt te relationsip between te effect and fixation effects sown in oter, earlier work in te general area of problem solving (Duncker, 1945; Lucins and Lucins, 1957). Given te potential significance of tis effect in design education and practice, in part of teir extension of tis researc Purcell and Gero (1991) used pictorial representations of a number of different possible design solutions to a single problem. One of te Jansson and Smit problems, te design of a bicycle rack for a car, was used, wit different groups being sown one of te possible design solutions. One of te pictured designs ad no effect on te designs produced, one of te examples was associated wit an increase in te number of designs, and one of te examples was associated wit wat appeared to be a fixation effect wit aspects of te example being produced more frequently tan in te control group wo were given only a verbal statement of te problem. Altoug tis result can be seen as indicating tat it is not simply te presentation of a pictured example wic produces design fixation, tere were a number of oter aspects of te experiment wic could ave influenced te result. Altoug eac of te pictured examples was based on an existing design, tere were wide differences between te examples in terms of te numbers of eac of te designs sold. Tis difference was replicated in judgments of te familiarity of te different examples made by te participants after tey ad completed te design session. Te example wic ad no effect on te designs produced was infrequently purcased and judged as unfamiliar by te participants, wereas te example wic was associated wit fixation-like effects was te most frequently purcased type and was judged to be igly familiar by te participants. As a result, te effect obtained could reflect te use of existing knowledge (wit te picture triggering reminding) about te pysical caracteristics of te example, based on frequent experience wit te example represented (furter weakening support for a fixation effect). In addition, tere were oter differences between te two experiments wic could account for te differences in te results. Te Jansson and Smit experiments ad involved advanced undergraduate and practising mecanical engineering designers, wereas Gero and Purcell ad used arcitectural and industrial design students at te beginning of teir design education. Te Gero and Purcell experiment,
Fixation effects: do tey exist in design problem solving? 335 terefore, differed in terms of te design discipline involved and te level of experience of te designers. Te aim of te experiment reported ere was to assess te possible role of tese differences between te two experiments in producing te inconsistent results by (1) using designers wit te same background (mecanical engineering) as in te Jansson and Smit experiment togeter wit a group wit different design backgrounds (for example, industrial designers); (2) using a problem were te likeliood of prior experience (on te part of te participating designers) bot of te problem and of existing design solutions is low; (3) using a Jansson and Smit example design togeter wit anoter design for te same object as te fixation material. Metod Selection of te design problem One of Jansson and Smit's (1991) problems involved te design of a device for use by te blind to measure quantities in cooking. Altoug suc devices do exist, tey would not be frequently experienced as a part of everyday life or as a part of an individual's design experience. Te design example from te original Jansson and Smit experiment was used, togeter wit a drawing of a quite different design based on an example obtained from te Australian Royal Blind Society (RBS). Te Jansson and Smit example [see figure Al(c) in appendix] consists of a compartmentalised box wit a carrying andle on top and a slide on te bottom wic is witdrawn in order to release te contents of te compartments. Te compartments are of equal size. Te basic operating procedure is to fill all of te compartments to te brim and ten, olding te device over te receptacle, to release te contents from te number of compartments required to make up te desired volume by pulling on te slide. To facilitate tis action (for blind persons) te slide produces audible clicks as eac compartment is emptied. Te design of te RBS example [see figure Al(d)] consists of a clear plastic cup wit four large graduation marks on te inside. Eac of tese is labelled wit raised Arabic numerals to indicate te proportion of a cup wic corresponds to a graduation. Te cup as a broad lip to facilitate pouring in any direction. Te presumed metod of operation for tis cup is to place it on a flat surface, find te appropriate graduation mark, place one's finger on it, and commence filling te cup wit a liquid or powder until it can be felt by te finger on te graduation mark. Te cup can ten be emptied into te intended receptacle wit one and, leaving te oter and free to assist in locating te receptacle. Te broad relatively ig brim lessens spillage problems during te dispensing operation. Procedure Te basic metodology used was te same as in previous design-fixation experiments (Jansson and Smit, 1991; Purcell and Gero, 1991). Te participating designers came from four distinct populations (described in te next section). Eac of tese populations was arbitrarily divided into between two and tree separate experimental groups (see table 1). In all cases one of te groups was designated as te control group ( c ) and were given a simple (verbal) description of te problem. Te oter (experimental) groups ( j for te Jansson and Smit example and R for te RBS cup example) were given te same description of te design problem togeter wit a picture of te relevant example. For tose receiving a pictured example, te verbal instructions included a single additional sentence drawing attention to te pictured example as an illustration of te level of detail wit wic designs were to be presented. Te facing pages of te (A4-sized) instruction seets used for te control
336 A T Purcell, P Williams, J S Gero, B Colbron and oter groups are replicated in figures Al(a) and Al(b), respectively. Te reverse pages of te instruction seets used for te Jansson and Smit example and RBS example groups are in figures A 1(c) and A 1(d), respectively. Te reverse page of te instruction seets given to te control groups was blank. Participants were randomly allocated to an experimental or control group and te design session lasted for 45 minutes. Participants were supplied wit seets of A3 paper on wic to do teir designs. Table 1. Allocation of subjects to experimental conditions. Population description Symbol Number in group 3 Total Tird-year mecanical engineers (j> x Fourt-year industrial designers (j> 2 Fourt-year interior designers ^3 Fourt-year mecanical engineers <f> 4 Total 0 c 15 8 4 8 35 10 7 7. 7 31 R 12 7 6 0 25 37 22 17 15 91 a c control group; j group given Jansson and Smit example; R group given Royal Blind Society example. Participants Tird-year and fourt-year undergraduate designers from mecanical engineering participated to equate bot te design background and te level of experience of te designers to tose in te original Jansson and Smit experiments. In addition to using designers from te same design discipline and wit a similar level of expertise as in te Jansson and Smit experiments, fourt-year undergraduates from industrial design and interior design participated to provide a test of te generality of te effect across design disciplines. Industrial designers were cosen rater tan, for example, arcitectural designers because te size and scale of te problem being attempted would be quite atypical for arcitectural designers but typical bot for industrial and for mecanical engineering designers. Te interior designers were included as an opportunity arose wic provided tem as anoter design discipline for comparison. Te total numbers of participants from eac population and teir allocation to experimental and control groups are tabulated in table 1 togeter wit te symbols used to represent eac population. Measures of design fixation In teir series of experiments, Jansson and Smit (1991) used two related types of measures of design fixation. Te first involved direct pysical similarities between te pictured design and a design solution and te second te reproduction of design flaws or mistakes present in te pictured example. Pysical similarities between designs can take a number of forms. Peraps te simplest involves te reproduction of te pictured design; tat is, te direct copying of te example design. Jansson and Smit appeared to be assessing tis type of similarity in teir measure wic involved judges assessing te overall similarity of te designs. Anoter type of similarity can involve te reproduction of elements of te design: tat is, parts or configurations of parts can be recognised in te designs produced. A tird type of similarity can be termed analogical similarity. Here te same design principles as te pictured example are used in te design witout te pysical form being te same. Te measurement of flaws in te design also as a number of aspects. Jansson and Smit (1991) assessed wat tey referred to as intentional design flaws: tat is, flaws deliberately introduced into teir example designs. However, analysis of teir
Fixation effects: do tey exist in design problem solving? 337 example designs in terms of te requirements for te design problem reveals a number of unintended design flaws. For example, in te design problem used in tis experiment te design of a measuring cup for use by visually andicapped individuals te intentional design flaws were tat only discrete quantities could be measured were te description of te design called for continuously variable quantity selection and tat no provision was made in te device for overflow. Te unintended design flaws are te following. (1) Te device [see figure Al(c)] requires two-anded operation and deprives te blind person of a prime sensor (touc) wen attempting to position te device over te receptacle into wic te contents are to be dispensed. (2) Te action of pulling te slide is igly likely to result in te movement of te wole device, exacerbating te problem of positioning te device over te receptacle. (3) Te long array of openings increases te criticality of positioning te device over te receptacle. (4) Te requirement to fill te compartments to te brim, in addition to creating te requirement for an overflow system, introduces te problem of maintaining te device level during operation, wic would be difficult for a blind person in spite of te lateral stability, because of pendulosity. (5) Te action of pulling te slide will exacerbate te balance problem. (6) Te only way to determine wen te compartments are full is to test for overflow by direct use of a finger. (7) Te device is bulky and awkward to use. Tere is also anoter problem wit Jansson and Smit's intentional design flaws. For example, not providing for overflow is an error of omission and it is difficult to claim tat flaws are being transferred into designs just because te same omissions occur. Rater, it is necessary to demonstrate te carry-over of committed errors, tat is, design flaws actually present in te pictured example suc as te unintended errors discussed above. A similar analysis can be made of te oter example design used in tis experiment. Te metod of filling te device to te required level as te undesirable requirement of using te finger as a direct sensor to detect te level of te contents. In addition powders ave a tendency to form piles wen poured and tis would cause te sensation of te powder on te finger to be an inaccurate indication of te quantity of te powder in te cup. [Tis problem is also present (to some extent) in te Jansson and Smit example.] Arabic numerals are unlikely to be easy for blind people to read wit teir fingertips even wen raised and te provision of Braille labels would be more useful. In common wit te Jansson and Smit (1991) design, tis design allows only te measurement of discrete quantities. From tis discussion tree sets of measures of design fixation are possiblemeasures of similarity, measures of intentional design flaws, and measures of unintended design flaws. Te specific implementations of tese measures used in tis experiment are described below and tabulated in table 2 togeter wit te symbols used to represent tem in various figures. Measures of similarity Te measures of similarity used were (1) judgments of direct pysical similarity of te designs to te Jansson and Smit example (J s, design feature 4 in table 2) and te RBS cup (JR S, design feature 15); (2) judgments of te analogical similarity of te designs to te Jansson and Smit example (/ a, design feature 5) and te RBS cup (R a, design feature 16). Te symbols for tese features are given in column 3 of table 2.
338 AT Purcell, P Williams, J S Gero, B Colbron Intentional design flaws For te Jansson and Smit example tese were (1) a design tat makes no provision for overflow (/*, design feature 3); (2) a design tat makes no provision for overflow, conditional on te design being inerently reliant on overflow in its normal operation (/ 2 8, design feature 7), obtained by testing for design feature 3 on condition tat design feature 1 is already satisfied; (3) a design tat only measures discrete quantities (/ 3 8, design feature 12). Te symbols for tese features are listed in column 4 of table 2. No known deliberate design flaws were present in te RBS design. Unintended design flaws For te Jansson and Smit example, tese were (1) a design tat contains a large array of output openings (J 3, design feature 8); (2) a design tat requires te use of bot ands to operate it (/ 4, design feature 9); (3) a design tat requires pulling, pusing, or twisting part of te device during te dispensing action (J 5, design feature 10); (4) a design tat requires tat te device be filled to te brim for proper use (J 6, design feature 11); (5) a design tat only allows te selection of discrete rater tan continuous quantities (/ 7, design feature 12). Teir symbols are listed in column 5 of table 2. In tis set tere were two furter conditional measures (6) a design tat requires te direct use of te finger to test for overflow were te design inerently depends on overflow (J l9 design feature 6); (7) a design tat makes no provision for overflow were overflow is part of te design (J 2, design feature 7). Teir symbols are also listed in column 5 of table 2. For te RBS example, te parallel measures were (1) te design only allows te measurement of discrete rater tan continuous Table 2. Design features. Design feature Uses overflow as part of design operation Direct use of finger to test for overflow Makes no provision for overflow Superficial Jansson and Smit similarity Analogical Jansson and Smit similarity Direct use of finger to test for overflow (conditional on 1) Makes no provision for overflow (conditional on 1) Large array of output openings Two-anded dispensing action Need to pull, pus, or twist during dispensing action Need to fill te device to te brim Only measures discrete quantities Makes use of non-braille labels Direct use of finger to sense te level of te substance Superficial RBS (Royal Blind Society) cup similarity Analogical RBS cup similarity Number Symbol(s) a R, a a, measures of similarity; b, intentional design flaws; c, unintended design flaws (Jansson and Smit example); d, unintended design flaws (RBS example). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 a J. R, b A 6 J$ /* c A d R, R, R,
Fixation effects: do tey exist in design problem solving? 339 quantities {R l7 design feature 12 and sared wit te Jansson and Smit example design); (2) te design uses raised Arabic numerals or Roman letters, rater tan Braille, to identify quantities [R 2, design feature 13); (3) te design requires te use of a finger as a direct sensor wen filling te cup to te desired level (R 3, design feature 14). Teir symbols are listed in column 6 of table 2. Tree judges were used to score eac of te designs independently for design features 1-5 and 8-16, wit te judges ten examining areas of disagreement by jointly assessing eac design and reacing agreement on te scoring for te relevant design features. Design features 6 and 7 were constructed by te conditional combination of features 2 and 1, and 3 and 1, respectively. Design features 1 and 2 exist purely for tis purpose. At te time of judgment te judges were not aware of eiter te population or te experimental group to wic te designer of te design belonged. Analysis and results Altoug te basic data consist of te number of occurrences of eac design feature in te designs for te different experimental conditions, tere are a number of ways in wic tese data can be treated. For example, te data can be converted to proportions of designs containing a given design feature. In addition, because participants were not limited to producing only one design, data can be extracted for te numbers or proportions of first designs from eac designer wic contain a given design feature, te numbers or proportions of designers tat produced at least one design containing a given design feature, or for te average number or proportion of designs for eac participant wic contain a given design feature. Altoug tese are te various possibilities available for tis data set, te focus in tis paper is on te possible reasons wy differences were found between te Purcell and Gero (1991) experiments and te original Jansson and Smit (1991) work. Te form of analysis wic is most similar to te original Jansson and Smit experiments is to analyse te number or proportion of te designs sowing a particular design feature and tis was te measure adopted. Te symbol N PX is used to designate te number of designs, taken from designers in te population P and allocated to experimental condition X, sowing a particular design feature, and \A\ PX is te total number of designs produced by designers in population P and experimental condition X. Te proportion of designs in te sample exibiting te particular design feature is ten given by N PX /\A\ PX. Te design features included in an analysis ave also been grouped in terms of te four sets of design features tat address te different issues discussed in te preceding section. Te measure, t, of statistical distance between two proportion estimates p 1 and p 2 used ere is given by t(pupi) = {Pi-Pi) p(l-p)\ - + - n { n 2 were p 1 n 1 +p 2 n 2 P =, (2) n l + n 2 and n x and n 2 are te sample sizes used to determine te estimates p x and p 2, respectively. Tis equation is a reformulation of te equation given by Freund (1979, page 318) as te appropriate statistic for ypoteses concerning one-sided comparisons (suc as p x > p 2, or p x < p 2 ) by substituting p x = x 1 /n l, and p 2 = x 2 /n 2. 1/2 (1)
A T Purcell, P Williams, J S Gero, B Colbron Its sampling distribution is Student's ^-distribution wit n x + n 2-2 degrees of freedom. In te experiments presented ere te aim is to test te null ypotesis, p x = p 2, against te alternative ypotesis, p x > p 2. Te proportion p x can be said to be significantly greater tan p 2 wit (1-a) confidence if Q[t(p u p 2 ), n x +n 2-2] < a, were Q(x, v) is te probability tat a variable drawn randomly from a ^-distribution wit v degrees of freedom will be greater tan x. Superficial and analogical similarity Jansson and Smit (1991) ad found tat judgments of te similarity of te participants' designs to te pictured example were iger for te group sown te example design. In tis experiment two types of similarity were identified examples sowing features directly similar to te pysical caracteristics of te example design or examples sowing pysical features wic embodied te same principle as te example design but were te pysical realisation of te principle was different. Tese two types of similarity are referred to as superficial and analogical similarity, and in figure 1 we sow te comparison between te four groups sown te picture of te Jansson and Smit design and te relevant control group on tese two measures. Te probability associated wit te (signed) difference in te proportions being as large as tat observed is sown on te abscissa wit te participants from te different populations sown separately. None of te populations exibits significant differences between te control and experimental groups in terms of te superficial similarity measure altoug te comparison for te tird-year mecanical engineering students (population fa) approaces marginal significance. Tis result indicates tat tere was little effect of te pictured example in terms of te reproduction of overt pysical features of te example. However, bot engineering student groups (fa and fa) sown te example design produce very significantly more designs tat are analogically similar to te example tan teir respective control groups, clearly demonstrating a fixation effect. By contrast, te industrial design and interior design students (groups fa and fa) sown te example demonstrate only a marginally significant increase over teir respective control groups. In figure 2 we present te parallel set of results for te groups sown te RBS design and te relevant control groups. [Fourt-year engineering students (group fa) did not participate in tis part of te experiment.] In tis case tere are no significant differences wit te superficial or analogical similarity measures for any of te types of students. Tese analyses demonstrate, first, tat tere was no simple reproduction of te pysical attributes of example designs; tat is, tere appears to be no simple copying of te pictured design. ^ P =, v = 48, v = 28, v = 25, v = 34 N fu _ NP, Ml* 0.99 0.999 0.9999 0.99999 Jansson and Smit similarity Figure 1. Comparison of adjudged similarities / s superficial and J a analogous) to te Jansson and Smit example for te experimental groups sown te Jansson and Smit example (,) and teir respective control groups ( c ). (Te assignment of te groups is given in table 1.)
Fixation effects: do tey exist in design problem solving? 341 Second and peraps of more interest, is te clear demonstration of a fixation effect in terms of te use of te principles involved in te pictured design, but only wit te example design used by Jansson and Smit (1991) and wit marked differences in significance levels obtained between te two mecanical engineering populations (groups (j) x and <f> 4 ) and te industrial and interior designers (groups <f> 2 and (j) 3 ). 01 0.00001 0.0001 0.001 0.01 0.05 0.5 0.9 0.99 0.999 0.9999 0.99999 _ - l I 1 : «^ ^ 1 ^zr^ - - ^ ^ ^ - ^ f F ^- p = -B- P = * v = 43 5 2, v = 40 J v = 16 N, N» \AP.S* Ml, RBS cup similarity Figure 2. Comparison of adjudged similarities (R s superficial and R a analogous) to te RBS (Royal Blind Society) example for te experimental groups sown te RBS example ( R ) and teir respective control groups ( c ). (Te assignment of te groups is given in table 1.) Intentional faults in te Jansson and Smit design Figure 3 presents te results of te comparison between experimental and control groups for te two intentional faults introduced by Jansson and Smit into te design for eac designer population. As discussed previously, one of tese intentional faults te absence of provision for overflow can be assessed in two ways. It can be eiter assessed directly or by taking into account weter or not te design inerently relies on overflow for its normal operation te conditional measure. Jansson and Smit ad found design fixation wit teir intentional design flaws, altoug it is not clear weter te nonprovision of an overflow device feature was conditional on overflow being a normal part of te operation of te devices. Figure 3, owever, clearly sows tat in te current experiment it is only wit te conditional measure (/ 2 8 ) tat tere are significantly more occurrences present and, as wit te analogical similarity measure, it is te two engineering groups {<j) l and (j) A ) and not oi 0~ p = p = p = p = Nr.t Ml,, o 0.99 0.999 0.9999 0.99999 J? 7 Jl Intentional design flaw Figure 3. Comparison of occurrence of intentional faults (see table 2) for te experimental groups sown te Jansson and Smit example (,) and teir respective control groups ( c ). (Te assignment of te groups is given in table 1.) *1 *?. K V V V V = = = 48 28 25 34
342 A T Purcell, P Williams, J S Gero, B Colbron te industrial design or interior design groups {(j> 2 and <f> 3 ) wic sow te effect. Altoug te results sown in figure 3 for te feature / 3 8 indicate no significant difference for any of te populations, it sould be noted tat te significance Jansson and Smit obtained for tis feature was only marginally significant 0.019 (t = 2.109, v = 86). Faults of commission in te Jansson and Smit and RBS designs In figure 4 we illustrate te differences in proportions between te experimental and control groups for te seven design features wic represent commission errors in te Jansson and Smit design. For te industrial design (group <j> 2 ) and interior design students (group <f> 3 ) tere are no significant differences between te experimental and control groups for eac of te design features. Overall, terefore, tese groups do not demonstrate design fixation in te sense of producing commission errors in teir designs of te type present in te Jansson and Smit design. Tese results contrast wit te analysis of te two engineering student groups {(/> 1 and (j> A ). Here tere are very significantly more occurrences in te experimental tan in te control groups for bot engineering student groups on several design features and most of te oters sow marginal significance. As a result te two engineering student groups sow quite clear evidence of design fixation according to te quite stringent criterion of errors of commission wereas te industrial design and interior design students do not. -S" P = fa, v = 48 -f P = fa, v = 28 B- P = fa, v = 25 -*- P = fa, v = 34 \\A\,, 4i ' \A\,,J 0.99 0.999 0.9999 0.99999 7, J 2 J3 J4 J 5 J 6 J-j Jansson and Smit design flaw Figure 4. Comparison of occurrence of faults of commission (see table 2) for te experimental groups sown te Jansson and Smit example (,) and teir respective control groups ( c ). (Te assignment of te groups is given in table 1.) 0.00001 0.0001 0.001 ^~ P = fa, v = 43-4 P = fa, v = 40 -s- P = fa, v = 16 Ol 0.99 0.999 0.9999 0.99999 *1 ^2 R3 RBS cup design flaw Figure 5. Comparison of occurrence of faults of commission (see table 2) for te experimental groups sown te RBS (Royal Blind Society) example ( R ) and teir respective control groups ( c ). (Te assignment of te groups is given in table 1.)
Fixation effects: do tey exist in design problem solving? 343 A quite different pattern occurs, owever, wit te RBS design (figure 5). Here tere are no differences between experimental and control groups for any of te features and for any of te groups of designers. Clearly, design fixation is absent for tis example design. Conclusions In te introduction to tis paper it was argued tat te failure to replicate te Jansson and Smit (1991) design-fixation results in te Purcell and Gero (1991) experiment could ave been caused by tree factors te inexperience of te design students used, differences between te design disciplines of te participants in te two experiments, and familiarity wit examples of te problem troug everyday experience. In tis experiment participants were from te same discipline and at te same level of education as tose in te Jansson and Smit experiment. Te design problem selected was one of tose used in te original experiments, but it was also a problem were it was unlikely tat participants would ave seen an example. However, te problem was of a scale tat would be in te range of experience of te designers. In tis experiment clear evidence for design fixation was obtained under te conditions wic most closely replicate tose of Jansson and Smit (1991). Tis appears to indicate tat te results in Purcell and Gero (1991) could ave been caused by te use of designers from a different discipline, given tat te selection of te problem in tis experiment ensured tat it would be igly unlikely tat participants would ave previous experience wit examples of solutions to te design problem and tat te scale of te problem was appropriate to te discipline of te designers. However, in tis experiment, design fixation only occurred wit one of te example designs tat used by Jansson and Smit and for advanced student designers from a mecanical engineering background tat is, te same type of background as te participants in te Jansson and Smit experiment. Tis result poses a number of questions about te design-fixation effect. At first glance, tis result migt be interpreted as indicating tat design fixation is not of muc significance because it appears to occur under suc a restricted range of conditions. It would also be possible to argue tat te conditions were te effect does occur may simply represent a set of conditions were te participants respond to te demand caracteristics of te situation and tat te effect is not related to te design process at all. Tere are a number of aspects of te results, owever, wic argue strongly against tese interpretations. First, te effects were obtained consistently wit all tree types of measures similarity, and intentional and unintended design flaws. Furter, te significant effects found wit te similarity measures were not in terms of superficial or direct similarity but in terms of analogical similarity, and te significant effects found wit faults in te design were not only in terms of errors of omission but were also in terms of te more robust errors of commission. In addition, if te results were only associated wit te demand caracteristics of te situation as tey affected engineering students, it would be expected tat similar fixation effects would ave been obtained wit te RBS example design. On te basis of tese arguments, terefore, it would seem tat te effects are genuine fixation effects. Te key question, terefore, is wy does fixation occur wit te engineering students but only wit a particular example design and not at all wit te industrial design students and interior design students? Some possible answers become apparent if te two example designs are examined in terms of te principles involved in teir design and correspondingly in assessing design fixation. Te RBS design is essentially
344 A T Purcell, P Williams, J S Gero, B Colbron a single graduated container. By contrast, te Jansson and Smit design is more complex and involves a number of engineering principles wic could be tougt of as mecanical design principles, for example te dispensing action involves mecanically moving parts. As a result it could be ypotesised tat fixation effects occur wen a pictured example design contains representations in pysical form of principles wic are a part of te expertise of te designer involved. Conversely, fixation will not occur wen te pictured example does not contain tis type of information, eiter because te principles involved are not specific to a particular design discipline as in te case of te RBS design or do not form a part of te knowledge base of te discipline involved as was te case wit te industrial and interior designers. Tis ypotesis can be tested by te use of design problems were te example designs are based on principles wic do or do not matc te discipline of te designers. From tis perspective, fixation represents te application of principles wic are relevant to a particular type of expertise in a situation were application of tose principles is inappropriate for te problem being attempted. Tis in turn provides an additional focus for researc in te area by posing te question of wy designers fail to identify te requirements of te problem wic make te application of te principle inappropriate, in contrast to te question of wat it is about te pictured example wic produces fixation. Anoter possible explanation for wy te Jansson and Smit example elicited design fixation wereas te RBS cup did not could simply be te complexity of te example designs. Te Jansson and Smit example, being more complex, would require more attention to understand (on te part of te subjects) tan te RBS cup wic is quite banal. Te difference in fixation effect could be related to tis difference in attention. Tis ypotesis can be tested by te use of design problems of graded complexity. Tis explanation does not address te problem of te differences between te mecanical engineers and te industrial and interior designers. A possible reason for te differences between te various disciplines could be found in teir education. It is possible tat te training of mecanical engineers teaces tem design metods tat are susceptible to design fixation and tat of industrial and interior designers does not. Conversely, it is possible tat umans ave a predisposition to design fixation and te design education of industrial designers and interior designers elps tem to overcome it and tat of mecanical engineers does not. Tese ypoteses can be tested by te use of subjects from various levels of education witin te various disciplines. If fixation occurs in design, even under limited conditions, it as te potential to affect te way design is taugt. Design is often taugt largely by precedence in suc disciplines as arcitecture; wat role does fixation play ere? Tis is still an open researc question; it is, owever, an important one. Acknowledgments. Tis researc as been supported by a seed grant from te Key Centre of Design Quality and by a grant from te University of Sydney's University Researc Grant Sceme. References Duncker K, 1945, "On problem solving" Psycological Monograps 58(5, wole of issue 270) Freund J E, 1979 Modern Elementary Statistics 5t edition (Prentice-Hall, Englewood Cliffs, NJ) Jansson D G, Smit S M, 1991, "Design fixation" Design Studies 12(1) 3-11 Lucins A, Lucins E, 1957 Rigidity of Beavior (University of Oregon Press, Eugene, OR) Purcell A T, Gero J S, 1991, "Te effects of examples on te results of a design activity", in Artificial Intelligence in Design VI Ed. J S Gero (Butterwort-Heinemann, Oxford) pp 525-542
Fixation effects: do tey exist in design problem solving? 345 APPENDIX Design Exercise: A Volume Measuring Apparatus for Use in Cooking by te Blind Te aim of tis design exercise is to design a volume measuring apparatus for use in cooking by blind persons. Te measuring apparatus designed sould possess te following attributes: easy to operate usable wit bot powders and liquids prevent waste of food capable of measuring quantities from \ to 1 cup easy to clean inexpensive Detailed and accurate drawings are not required. Simple, roug outline sketces are all tat is needed. Te sketces may be annotated wit written comments as required to clarify your intentions. You will be allowed forty five minutes to complete te design. If you wis you may complete more tan one design. Design Exercise: A Volume Measuring Apparatus for Use in Cooking by te Blind Te aim of tis design exercise is to design a volume measuring apparatus for use in cooking by blind persons. Te measuring apparatus designed sould possess te following attributes: easy to operate usable wit bot powders and liquids prevent waste of food capable of measuring quantities from \ to 1 cup easy to clean inexpensive Detailed and accurate drawings are not required. Simple, roug outline sketces are all tat is needed. Te sketces may be annotated wit written comments as required to clarify your intentions. See figure 1 (on reverse page) for an example of te level of detail required. You will be allowed forty five minutes to complete te design. If you wis you may complete more tan one design. (b) equal sized compartments audible click for eac compartment 4 large (finger sized) graduation marks raised Arabic numeral labels Figure 1. Example Design Figure 1. Example Design (c) (d) Figure Al. Facing.page of instruction seet for participants in (a) control conditions, (b) oter conditions; and reverse page of instruction seet for participants in (c) Jansson and Smit example condition, (d) Royal Blind Society example condition. (Te reverse page for te control group is blank).