Empirical Study

Direct Manipulation Promotes Architects’ Productivity


Objectives: The aims of the present study are to verify whether direct manipulation ((hand gesture input) could be a potential method of boosting designer's productivity compared with indirect manipulation (mouse input) based on the moderating effects of designers’ working experience.

Design: The study is conducted by a true experiment using two independent variables (namely, modeling method x working experience) and one dependent variable (productivity)

Method: Eight architects with different working experience were asked to build certain basic 3D shapes digitally or physically either using hand/clay or by computer 3D modeling software in an architect company conference.

Result: According to the findings, by switching modeling method from computer-based to hand/clay-based method, the overall productivity in both senior and junior designer’s groups can be boosted. However, there is a modeling method x working experience interaction in which the effects of modeling method on productivity are dependent on designers working experience.

Conclusion: By discussing the observation from the experiment, this study draws the conclusion that direct manipulation could significantly boost the productivity if designer lacks working experience. However, the study here also has a few limitations which affect validity. The future work and experiment should aim at avoiding these affections.


Direct Manipulation, Gesture Modeling, Design Process, HCI, Productivity


Almost half the daily work of an architect is modeling. The 3D model could bring designers more comprehensive ideas concerned with the design project, yet the way designers make models matters to their work efficiency. This study assessed the moderating effects of direct manipulation on human’s productivity, especially in designer’s office.

Architectural design is virtually a process of materialization reflecting the creative thinking of designer (Lobel2009). The process of design refers to imagination interacting with the assistance of all types of tool. Tools selected by architects have significant effects on design development as well as the productivity of designers. Many architects are giving considerable efforts learning how to use software and solving software problems that are not the problems in design. The manipulation of the user interface, interacting through data, has been troubling professional architects, leading to the reduced efficiency and misunderstanding of projects.

The progress of architectural design depends largely on the assistant of computer-aided design tools which have made extraordinary contributions to architectural design during the past decades (Kalay 2004). However, current Computer-aided design tools are fundamentally confined into windows, icons, menus and pointer (WIMP) interface, including 2D based input manipulation and 2D output display. Building geometry, as a physically 3D object designed based on 2D based workflow, is bringing designers less interaction with the design object.

People always employ the indirect manipulation input to achieve the design intent. The limitation originates from both physical configuration and software setting. For physical configuration, keyboard is used for typing to input command, whereas what an architect is doing is communicating with the object to develop the design idea rather than giving command to control the object (Richards et al. 2009). What designer needs is a spatial platform to operate object wherein designers could have a multiple dimension reference. The mouse appears to be a spatial platform because of its x and y operating space, yet it just has two dimensions. The use of two-dimension manipulations to operating 3d space remains unmatured. For the software side, the WIMP interface is a combination of windows, icon, menus and pointing, according to Van Dam’s study. Each of these items represents a command or a purpose. Sometimes the user who calls one icon must find another one. To use the WIMP interface, designers must pay extra attention on software learning. In particular, some interfaces are not friendly enough for designer to learn and use.

Architects with typical background were trained and practiced in the means and methods of design. Design is an act of seeing, thinking and making, (Lobel 2009) combining the efforts of eyes, brain and hands. Architect is responsible for the aesthetic and spatial details of the building -- its size, shape, use of space and site requirements (Binkley 1997). This requires that regardless of the tool designer selected (either pencil, mouse, etc.), design aids should coordinate the interaction among eyes, brain and hands when designer is observing, thinking and making. Through the interaction of brain, hands and eyes, architects can be aware of information; through the repetitive creation, architects would be likely to re-define information and get feedback so that they can make the right decisions (Kalay 2004).

In such scenarios, the development of new design tool should consider embodying direct manipulation to boost the productivity of designers. The direct manipulation concepts can be interpreted by car driving based on manipulating the steering wheel and pedals. The car responds immediately to drivers’ actions, and these responses are immediately significant. If a driver is making a mistake, e.g., turning too sharply, he can quickly recognize this and take the correcting measure. Turning steering wheel to change the direction and stepping on gas to speed up are directly related. But when a passenger sits in the backseat of the car giving stranger directions. The indirect manipulation will lose the feeling of the road, and the passenger does not have a direct view of where he is going.

Shneiderman (1993) argued in his most frequently cited study that the most intuitive tool should incorporate direct manipulation via a natural language rather than not a programming language. The natural language in this context refers to not only an oral language like English but also the spontaneous behavior of human beings, e.g., hand gestures. If the manipulation setting could base on spontaneous behaviors, it will be easy to learn.

However, our understanding of human’s spontaneous behaviors remains in the early stage, in particular natural gesture. The exploration of natural manipulation aims to find how it could impact the process of design. Grandhi (2011) discussed the definition of natural and intuitive gestures. He argued that the concept of intuitive and natural gesture should be addressed by two research questions. 1) Which aspects of a gesture (e.g., motion, hand shape and form) are natural and intuitive when communicating a transitive action? Whether gestures using hand actions holding-imagined-objects are more intuitive and natural than those using body-part-as-object? 2) Whether it is easier to gesture a transitive action when the user communicates as how he/she habitually performs the action or when the user communicates it as an instruction? In other words, one gesture should be “this is how I do it” or “this is how you should do it”. According to Grandhi’s experiment, user’s experience could be enhanced by developing the gesture vocabulary based on understanding in which the actions are embodied. From the experiment, a guideline was developed for design of a touchless gestural interface.

Thus far, there have been few studies focusing on direct manipulation in designer oriented working environment. Fritz, et al. (2009) demonstrated the use of wireless motion-sensing objects (e.g., the Wii Remote) to control navigation of 3D architectural models. Francese et al. (2012) presented several projects using the Wii Remote and Kinect to navigate a map in a 3D digital environment. However, navigation as a single function will be insufficient for designers to design with spatial gesture. Hilliges et al. (2012) developed “HoloDesk,” a system that allows users to pick up, move and even shoot virtual 3D objects by their hands, in addition to which the system recognizes and responds to the presence of inanimate real-world objects, e.g., a piece of paper or an upturned cup. HoloDesk is a good attempt to change the way that people interact with a computer, yet it remains oriented to the needs of a regular user other than a designer. Other significant attempts that explored the possibility of designers to get support from gesture input are as follows. Yi, et al. (2009) presented a novel method of gesture modeling to support architectural design. They define various components of architecture based on a set of signs. However, their gesture recognition depends largely on the application of colored marker to the user’s hand and numerous motion capture devices. It is not a simple and inexpensive system. Gross and Kemp (2001) used video-based gesture recognition which captures the 3D with two simple orthogonal webcams. Their configuration requires a white background and the use of a black glove so that the location and shape of hand are accurately captured. It is relatively limiting users’ experience, and the system has few simple gestures, limiting the application of their gesture modeling system to simple functions, e.g., distorting a mesh or moving existing objects. Kumaragurubaran (2012) presented a hands-free prototype of gestural support for design using a Kinect to manipulate a Grasshopper (a Rhinoceros plugin) interface and parametric model. However, manipulating Grasshopper components is not directly interacting with digital geometry.


Compared with using input device, e.g., mouse and keyboard, to conduct a design project, it is assumed that direct manipulation could give designers more efficiency and less learning cost while doing their job.


Design: The true experiment will be performed in a professional design environment (Architect Company). The first independent variable is the way all participants will use to make models to 3D shapes, whether the use of computer 3D software or hands with clay. The second independent variable is the professional experience that all participants have. The dependent variable is the productivity that participants had during the study. The way of how participants make model is within-subject factors and professional experience is between-subject factor.

Participants: Participants are from a world leading Architect Design Company located in Seattle, WA. All 8 participants are varying with working experience and position level, designers with no over 4-year working experience will be marked as junior designer, the rest will be marked as senior designer. All 8 participants are familiar with 3D modeling software Rhinocero to be used in the experiment at different levels. There are 4 males and 4 females in the experiment with the average age of 31.

Procedure: First, participants will be randomly assigned a A4 sized picture of a set of 3D shapes and will be asked to make the first set of 3d massing by computer as much as they can within 10 min according to the picture they have.

Second, they will be randomly assigned a different A4 sized picture of a set of 3d shapes and will be asked to make another set of 3D massing by hand and clay as much as they can within 10 min according to the picture they have.

Setting: The study will be conducted in a 200 Sq feet conference room in the Design Company in Seattle. In the room, a round conference table with radius of 2 feet is provided. Participants will provide a modelling clay cube with measurement of 8”*8”*8”for clay modeling. Participants are asked to bring their own laptop for software modeling.

Measures: The dependent variable, productivity, will be measured by the number of 3d massings that participants make by both computer-based and hand/clay-based method. The qualified 3d massings should be similar or the same as the 3D shape provided in the experiment.


A 2 X 2 (Modeling method X working experience) factorial analysis of variables (see Table 1.) Performed on the dichotomous data (computer modeling or hand/clay modeling) resulted in highly significant results.

Main Effect

As shown in Table 1, the participants are assigned to computer modeling approach in the junior designer group reported an average score of 2 (SD= 0.81) on the productivity measure. And an average score of 4.75 (SD=1.25) with hand/clay modeling approach; participants assigned to computer modeling approach in the senior designer group score an average of 3(SD=0.81), and an average of 4(SD=0.81) in the hand/clay modeling approach. Mean score varies from 2 to 4.75 across the four scenarios (SD=0.81-1.25).

The results suggest that there is a main effect of designer’s modeling method on productivity. The average productivity for both groups is improved, and the productivity measured by hand/clay made models is significantly boosted from the productivity measured by computer-based models. Productivity of hand/clay modeling approach (mean=4.375) is greater than that of computer-based method (mean =2.5). There is another main effect of designer’s working experience on productivity. General speaking, designers who have longer working experience have higher productivity. The productivity measured in the senior designer group (mean= 3.5) is greater than junior designer group (mean=3.375).


The experiment result suggests that Designer’s modeling method X working experience has an interaction with designers’ productivity as shown in Fig. 1. Designer’s modeling method effects on designer’s productivity depended upon designers working experience. Specifically, the subjects who have less than 4-year working experience perform more efficiently (mean=4.75) when using direct manipulation modeling method with hand and clay compared with using computer (mean= 2 ) while senior designers are improved less significantly (mean=4) when switching modeling method from computer modeling to hand/clay modeling (mean=3).


This cross-sectional study verifies whether the direct manipulation can boost the productivity of designer with moderate effects of working experience in a professional working environment for design, compared with the traditional modeling method that mostly uses mouse input and 2D monitor display. Most of the previous studies that are from HCI background have demonstrated that direct manipulation has positive effects on professionals performance, efficiency and cognitive perception (Kieras2001Shneiderman1997) through either theoretical or empirical studies, whereas there has been no comprehensive study from the perspective of designers, especially architects. For those studies from the field of architectural design, the current literatures have been rarely conducted using an empirical approach.


According to the hypothesis above and based on this empirical study, the hypothesis is supported that the direct manipulation boosts the productivity of designers, especially among the younger designers. Design activity that based on natural behaviors (gestures), will be very easy to learn. The circular flow of information among thinking, observing and making brings designers the opportunity to dynamically re-define the objects they sense. Body manipulation motivates designers to assess object from multiple aspects, thereby inspiring appropriate design intent. Grandhi (2011) explored people’s natural gestures using “before” and “after” pictures and instructed participants to perform the gesture required to get from before to after. The experiment suggested that user experience could be improved by developing the gesture vocabulary based on understanding in which the actions are embodied. Gross (2001) reported that hand gestures are a type of communication between people. However, user still limits human-computer interaction to cumbersome mice movements. The use of hand gestures in the past of human-computer interaction has aroused huge attention in the past several years. Some relevant works have been done in some research projects. The early concepts of hand gesture recognition depended on the sensible gloves. Special glove-based devices have been developed to analyze finger and hand motion and exploit these motions to manipulate and explore virtual worlds since 1980s. To further enrich the naturalness of the interaction, different computer vision-based techniques have been applied. In the meantime, the need for more efficient systems has resulted in new gesture modeling approaches.

However, the experiment result exhibits that direct manipulation are affecting less on productivity among senior designers than junior designers was not expected. As mentioned at the beginning, almost half of an architect’s daily work is modeling, for those designers with longer experience of operating modeling software, are spending less time on familiarizing with software while contacting his/her work task. For senior designers, however, the more experience they have with modeling software, the more accustomed they become to operating in a visual environment rather than a physical environment. When comparing with junior designer group, both groups are newly exposed to physical modeling with hand/clay. As a result, the physical modeling has significantly positive effects on junior designers.

Nevertheless, Norman (1988) highlighted that direct manipulation has undeniable enhancement function, and psychology literature references the advantages of physical representations in learning speed and retention. Direct manipulation exploits these advantages, developing systems with simple and impressive operation. Since complex syntax does not have to be remembered, and analogical reasoning can be used, less errors are made. When they are made, they are easy to correct by reversible actions. Reversible actions also foster the action of exploration as the fear of breaking something has been diminished. Moreover, people can gain confidence and mastery as they are in control and the system responses are predictable and immediate.


The experiment performed here contains few limitations that should be evoked and re-considered in the future experiment.

For external validity, there are negative affection of selection x treatment. The participants are selected from a world leading Architect Company primarily designing large scale commercial buildings across Asia, North America, and Middle East. Most of the projects are considered as landmark in the local market, and geometries of the buildings are relatively more complicated. The employees selected to participate in the experiment are equipped with higher modeling skill than employees from regular architectural design firms. Even though all participants are randomly selected from whole company, the average levels of professional skills (e.g., modeling) are considerably higher than those of other firms. If the experiment is performed in a designer firm primarily dealing with simpler geometric architecture, the result could be different. To avoid this negative effect, the participant selection should aim at larger pool, e.g., the architects registered in the state of Washington but not from a single company. Another negative effect on external validity is setting x treatment. The overall setting of the experiment locates in a separate meeting room with a round table. The environment is relatively independent from other interference source. However, in a real working environment, designers should address multiple interference, e.g., open space, noise and light conditions. The results in a practical working environment might be different with the separated room. Moreover, architects or designers’ daily jobs always require a team work, and the productivity may be associated with the basic workflow of the. The result in a real working environment associated with different individuals could be different with the experiment here, a solo work basis task. To avoid the negative effects on the setting X treatment, all participants should be introduced into a real working environment.

For internal validity, the experiment could be affected by the testing. The participants have the same pretest and post-test. All participants use the same examples sheet for the test. The examples they were using for the computer-based test are the same to those they were using for the hand/clay-based test. By viewing the examples twice, participants could be more familiarized with the topological relationship of all shapes. It is likely that participants performed better in the hand/clay-based modeling session than the computer-based modeling session. To avoid such negative effect, it is highly recommended to use different examples over the pretest and post-test. There was another negative effect on the internal validity which is compensatory rivalry. Four junior designers participated in the experiment, two of which were newly graduated from school. In the experiment, they showed great enthusiasm and ambition. During the computer-based modeling session, they tried to show off their modeling skill and the performance was a bit over average level. To avoid such negative effect, not letting participants know what this experiment is measuring may be helpful. Low reliability of measure affects statistics validity. The standard deviation suggested that the score of junior designer group are a bit more unstable when switching modeling method from computer-based to hand/clay-based method. When performing the hand/clay session with junior group, a few designers found the direct manipulation is relatively intuitive and fun to operate, and it was reported that he did not stop right away when time was up. Thus, the result for that sample is a bit over average level. To avoid the negative effect on statistics validity, all participants should abide by the experiment instruction. The study is also affected by the small size and lower power. The sample size (N=8) is too small to support further study in the future. As selected from Architecture Company, 8 participants are not sufficient to address the full scope of architect’s work. When 8 participants fall into two different groups based on working experience, the power of this independent variable will be lower than expected, making the result incomprehensive.

For construct validity, there might be a negative effect of hypothesis guessing. Before conducting experiment and during the recruiting sessions, the author explained to all participants that what this experiment aims at and what were the hypothesis of the experiment. In that case, participants were knowing what the author is expecting. It would somehow impact the final performance at some level. To avoid the negative effects of hypothesis guessing, the best solution will be that all participant should not be aware of what intervention has been performed throughout the experiment. There is another negative effect of poor construct definition, the study was aiming at measuring the productivity as outcome and dependent variable. Productivity is measured by the number of model made by participants. However, productivity contains wider concept and can be defined with more aspects. For instance, productivity can be also measured by the resource cost during the production process. Only measuring the number of outcome is not enough to assess the productivity of designer. If a designer requires relatively more clay or more expensive hardware for his/ her work, even his/ her outcome situating in a higher level, it will be hard to claim this designer has higher productivity.

Future Work

First, through the experiment performed, the hypothesis is supported within certain groups that whether modeling method has positive effects on designer’s productivity depends upon the moderate effects of working experience, yet as mentioned above, the negative effects on all types of validity should be evoked and re-addressed in the future study: the participant’s selection should be gathered from a wider range rather than from one single company. Also, the test environment should be closer to designer’s daily working environment. The experiment design and arrangement should be re-considered since using the same example twice could bring participants opportunity to get familiar with the shapes. In the meantime, a larger sample should be used to re-assess the hypothesis since the pool of the study is not sufficiently large to cover the full scope of architect’s work. Furthermore, participants should not be told what the purpose of the study, and their working cost will be considered into the measurement.

Second, the theoretical foundation also should be enhanced. Given the significance of visual feedback to designers, visual immersion incorporating both geometric and non-geometric feedback is desirable, yet existing hardware often interferes with UI interaction or communication. An augmented design system which should combine both visual and physical working environment/ indirect and direct manipulation could help designers enhance the experience of design and achieve the desired results. The future study related to this subject will aim to develop a new multidimensional computational design tools incorporating both 2D graphic user interface into 3D tangible working environment to facilitate designer’s interaction with the object and reduce learning time. Space of designer oriented multidimensional computational tool still needs further studies. The present study aims to establish an understanding of the space and develop empirical foundation which could embody the vision the author seeks to explore. The invention of designer oriented multidimensional computational tool will be likely to change the traditional workflow of designers thoroughly and help them gain new working experience.


Asanowicz, A. (1999, September). Evolution of Computer Aided Design: three generations of CAD. In Proceedings of the 17th conference of education in computer aided architectural design in Europe, Liverpool.

Ackermann, E. K. (2004). Constructing knowledge and transforming the world. A learning zone of one's own: Sharing representations and flow in collaborative learning environments, 1, 15-37.

Binkley, T. (1997). The vitality of digital creation. The journal of aesthetics and art criticism, 55(2), 107-116.

Van Dam, A. (1997). Post-WIMP user interfaces. Communications of the ACM, 40(2), 63-67.

Grandhi, S. A., Joue, G., & Mittelberg, I. (2011, May). Understanding naturalness and intuitiveness in gesture production: insights for touchless gestural interfaces. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 821-824). ACM.

Gross, M. D., and A. J. Kemp. (2001) “Gesture Modelling: Using Video to Capture Freehand Modeling Commands” in Proceedings of CAAD Futures 2001, 271-284. Eindhoven: Kluwer Academic Publishers

Hilliges, O., D. Kim, S.Izadi, M. Weiss, and A. D. Wilson (2012) HoloDesk: Direct 3D Interactions with a Situated See-Through Display. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems: 2421-2420.

Kalay, Y. E. (2004). Architecture's new media: Principles, theories, and methods of computer-aided design. MIT Press.

Kumaragurubaran, V. (2012) “A Multi-user Collaborative Space for Architectural Design Reviews” in Proceedings of the 2012 ACM annual conference extended abstracts on Human Factors in Computing Systems Extended Abstracts, 1381-1386.

Lobel, J. (2009) “Computer Aids Design” in Computational Constructs: (pp.104-112). Cambridge, The MIT Press.

Mart, D. C. (1982). A computational investigation into the human representation and processing of visual information. Free-man, San Francisco, CA.

Norman, D. (1988), The Psychology of Everyday Things, Basic Books, Inc.

Shneiderman, B. (1993). 1.1 direct manipulation: astep beyond programming languages. Sparks of innovation in human-computer interaction, 17.

Stiny, G. (2006). Shape: talking about seeing and doing. MIT Press.

Sutherland, I. (1975). Structure in drawings and the hidden-surface problem. Reflections on Computer Aids to Design and Architecture, 73-85.

Yi, X., S. Qin, J. Kang (2009) “Generating 3D architectural models based on hand motion and gesture” in Computers in Industry. 01/2009; 60: 677-685. Amsterdam: Elsevier B.V.