Knowledge workers’ stated preferences for important characteristics of activity-based workspaces

ABSTRACT

It is important for organizations to support employees with suitable workspaces during different activities. Increasingly the activity-based office is seen as the optimal solution, as it provides people with choice between different types of workspaces to perform specific activities. This study addresses the trade-offs knowledge workers make when choosing a workspace for performing three different categories of activities (individual concentration work, informal interactions, and formal interactions). A stated choice experiment was performed through a questionnaire among 251 Dutch knowledge workers in 14 organizations, measuring preferences for hypothetical workspaces described by their six most relevant aspects. Multinomial logit models were estimated to identify the workspace preferences during each activity. Additionally, Latent class models were estimated to find possible segments of workers with specific preferences that need to be supported differently by workplace managers. Findings showed that preferences for psychosocial design aspects (noise, workspace enclosure and control) were more important for workspace selection than indoor environmental quality aspects. They also show more differences between specific activities. Also, not everybody finds full enclosure important for individual concentrated work, and relatively older employees appear to be more critical about workspace suitability to support both formal and informal interaction.

KEYWORDS:

• Occupant surveys
• workplace effectiveness
• workplace design
• decision-making
• stated choice

Introduction

Office design that supports knowledge worker activities and workplace experience has been proven to improve individual (perceived) productivity (Al Horr et al., 2016; De Been & Beijer, 2014; Xuan, 2018). Nonetheless, several studies claim that knowledge-intensive organizations are currently not paying enough attention to adapt the environment to the activities of their knowledge workers (e.g. Chadburn, Smith, & Milan, 2017; Feige, Wallbaum, Janser, & Windlinger, 2013). Such disregard could harm organizational goals, because cost savings in buildings should not come at the expense of the employees’ ability to do their work; since staff costs are much higher than building costs (Newsham, Veitch, & Hu, 2018). Poor office buildings can result in occupants experiencing Sick Building Syndrome symptoms, such as headaches, exhaustion, inability to concentrate, and reduced work efficiency (Joshi, 2008; Redlich, Sparer, & Cullen, 1997), which is disastrous for a company’s competitiveness. Less productive employees can get less work done, have a lower contribution in meetings, create less value for organizations and are in that respect more expensive (Mawson & Johnson, 2014).

With the rapid implementation of the so-called activity-based office (ABO) concept in many countries (63% of large enterprises had already rolled out or considered it in 2011; see Dixon & Ross, 2011), organizations aim to improve workplace support of different activities. In ABOs, people no longer have dedicated desks and share different types of available workspaces to choose from for each of their activities when they are in the office (Rolfö, Eklund, & Jahncke, 2018). So, they obtain more autonomy in finding the proper support of their needs and preferences for different types of activities. There is however very little research on how employees use ABOs. The few existing studies show that they are not used as intended, because many employees tend to stick to one workspace and do not switch often; even when another space would fit their next activity better (e.g. Hoendervanger, De Been, Van Yperen, Mobach, & Albers, 2016). Research to understand whether, and how, an ABO can be perceived as supportive for employees is lacking (Gerdenitsch, Korunka, & Hertel, 2018).

Therefore, this paper studies the trade-offs that knowledge workers make based on their stated preferences for important ABO workspace design characteristics, when choosing between different types of hypothetical ABO workspaces for specific categories of activities (individual concentration work, informal interactions, and formal interactions). Because it is difficult to control and measure the complex and intertwined set of workspace variables in real life and at the same time (Kamaruzzaman et al., 2018), this study takes a different approach than most existing workplace studies by performing a stated choice experiment with 251 Dutch knowledge workers from 14 organizations. Participants in the experiment are asked to indicate their choice between 9 sets of 3 systematically varied ABO workspace descriptions. Based on these choices, a multinomial logit model can be estimated that shows the perceived relative importance of workspace characteristics for knowledge workers in order to be productive during specific categories of activities. Insight in which characteristics they find most important when choosing a specific workspace for a specific activity can be used as input for more user-focused ABO workspace design.

There are well-known differences in individual workplace needs and abilities (see e.g. Kristof-Brown, Zimmerman, & Johnson, 2005), for example regarding age and education level (Haynes, Suckley, & Nunnington, 2017) or gender (Lai & Yik, 2007). But as De Been, Van der Voordt, and Haynes (2016) stated, more research still needs to be conducted ‘ … in understanding individual needs and preferences of different groups … ’ (p. 10), because it is very likely that different types of employees with different preferences work within a single company. Therefore, also Latent class analyses are estimated on the obtained stated choice data, to check for possible segments of knowledge workers with similar specific preferences.

To summarize, the main aim of this study is to identify knowledge workers’ preferences for important ABO workspace characteristics to execute certain activities while in the office. A secondary aim is to check whether clear groups (segments) can be defined based on these preferences. Such insights are valuable when workplace managers need to justify choosing between investment options during work environment design/change projects before they are actually implemented. This research provides the workplace research field an example of the application of a methodology that makes it possible to study preferences for not yet existing or not available alternatives in real offices, instead of the common Post Occupancy Evaluation (POE) approach (e.g. Al Horr et al., 2016) or experiments in controlled laboratory settings (e.g. Kamaruzzaman et al., 2018).

Knowledge worker activities

Generally, knowledge workers can be defined through their activities, which are focused on thinking, problem-solving, collaborating and networking (Reinhardt, Schmidt, Sloep, & Drachsler, 2011). Through an extensive literature review of knowledge management literature, Reinhardt et al. distilled a list with as many as thirteen different knowledge worker actions ranging from (co-)authoring, analysing and monitoring to gathering information, learning and networking. However, their study concentrated only on individual tasks of knowledge workers, performed while sitting behind a computer.

As De Been et al. (2016) mentioned, the activities that people consider most significant for their productivity in the office is support of both concentration and communication. So, besides the diverse actions of individual concentrated work, it is also very important to support face-to-face interactions at work. Especially tacit knowledge is shared best during face-to-face interactions (Coradi, Heinzen, & Boutellier, 2015) and spontaneous face-to-face interactions are said to be most effective for engaging in creative or innovative work (Ngah & Jusoff, 2009). Both informal and formal interactions are important for knowledge workers (Marouf, 2007). Some studies show that people even share more knowledge during informal interactions than during formal meetings (Wensley 1998). Most of these often spontaneous interactions take place at or near the workspace and not in meeting rooms (Appel-Meulenbroek, de Vries, & Weggeman, 2017). Knowledge management researchers have shown a lot of different actions during interactions, such as reporting, demonstrating, asking questions, warning, etc. (see Berends, 2005).

As the extensive lists of actions would be too detailed and laborious to study with a stated choice experiment, only the three main categories of knowledge worker activities are used in this paper: individual concentrated work, formal interactions and informal interactions (De Been et al., 2016).

Important workspace characteristics in activity-based offices

ABOs generally have open-plan layouts, intended to increase the interaction between employees from different departments through serendipitous meetings, plus many places for formal and informal meetings. But small concentration cells and the possibility to work from home are also offered, because openness and communication can be a distraction for those employees trying to concentrate. This section reviews studies on how different workspace characteristics influence employee productivity (during activities in the office) and other important outcomes, in order to select the most important characteristics to identify employee preferences for. Some of these studies try to identify direct effects of specific design characteristics (e.g. Vimalanathan & Babu, 2014), while others focus on an indirect effect through comfort levels and/or workplace satisfaction (e.g. Mulville, Callaghan, & Isaac, 2016; Roulet et al., 2006).

According to a systematic review of empirical studies on possible employee outcomes from workplace design, Appel-Meulenbroek, Clippard, and Pfnür. (2018) found that most studies include psychosocial characteristics on a building level such as personal control and privacy. On the workspace level, they also identified many studies proving effects of the indoor environmental quality (IEQ), the individual space and its quality (ergonomics), and general décor and window views in the office, so these categories of characteristics are discussed below.

Regarding important psychosocial characteristics, privacy and having the ability to work distraction free is one of the strongest predictors for perceived individual productivity (Candido, Kim, de Dear, & Thomas, 2016; Leaman & Bordass, 2000). People need a certain feeling of anonymity and do not like the feeling of always being exposed, which is closely related to the level of enclosure (Clements-Croome, 2006). Besides, noise can decrease productivity rates by three to seven percent (e.g. Hongisto, 2005; Witterseh, Wyon, & Clausen, 2004). Additionally, personal control over a range of office features, such as temperature, ventilation, windows or the ability to personalize the workspace, show a strong link with enhanced work performance, comfort and also the acceptability of the workspace (Brager & De Dear, 1998; DeRango & Franzini, 2002; Kroner, Stark-Martin, & Willemain, 1992).

The IEQ has also been studied a lot in relation to employee outcomes such as productivity, satisfaction or comfort (e.g. Kim & De Dear, 2013). This includes lighting, indoor air quality and thermal comfort (Appel-Meulenbroek et al., 2018). Its effects are proven in various laboratory settings (e.g. Roulet et al., 2006) and other types of studies (e.g. Hameed & Amjad, 2009). Lighting has been proven to relieve occupants’ eye symptoms and tiredness, decrease motivational problems, and improve productivity (Kang, Ou, & Mak, 2017). Better air quality results in less Sick Building Syndrome symptoms (Niemelä, Seppänen, Korhonen, & Reijula, 2006), lower frequency of breaks (Mulville et al. 2016), less eye irritation (Wolkoff, Wilkins, Clausen, & Nielsen, 2006), happier and healthier workers (Hameed & Amjad, 2009) and a slightly higher productivity (Feige et al., 2013). For indoor climate the optimum differs substantially between individuals, due to varying preferences and personal characteristics; for instance, female employees generally are more dissatisfied with climate comfort (Karjalainen, 2012).

Regarding the individual space and its quality, the importance of ergonomic adjustable furniture should not be underestimated in its influence on employee productivity (Elbert, Kroemer, & Hoffman, 2018; Hameed & Amjad, 2009). Ergonomics means providing ergonomically designed workspaces that support employees’ job tasks, and enhances their health, well-being and ability to adjust their workspace to changing work processes (Robertson et al., 2009).

There are some studies on the general décor and window views in offices, but the evidence of effects on employee outcomes is not yet as solid as for the above characteristics. For example, Moya, Van den Dobbelsteen, Ottelé, and & Bluyssen (2019) reviewed the literature on effects of plants on people and indoor climate, but concluded that the full capacity of plants still needs to be clarified. There are also some studies providing first proof of increased work ability and job satisfaction with natural views (e.g. Lottrup, Stigsdotter, Meilby, & Claudi, 2015), and a combined effect of colour, texture and the shapes of interior design that leads to the overall feeling of well-being (Al Horr et al., 2016).

Summarizing, the following workspace characteristics appear to have a (slightly) proven relation with employee outcomes: psychosocial (privacy, level of enclosure, personal control and noise), IEQ (temperature, air quality, and lighting), and individual workspace quality (ergonomics, interior design). The next section explains the further research approach and which of these characteristics could be taken up in the empirical part of the study.

Research approach

Data collection

Existing studies on the importance of workspace characteristics for employee outcomes often use either AHP (Analytic Hierarchical Process) to obtain the opinion of building experts (e.g. Lai & Yik, 2007) or POE (Post-Occupancy Evaluation) questionnaires to identify the perceived importance by end users (e.g. Vischer, 2009). This paper takes a different approach by using specified hypothetical workspace profiles based on a stated choice experiment (see Hensher, Rose, & Greene, 2015), to rank the importance of different workspace characteristics based on trade-off decisions. An advantage of this method is that the characteristics cannot only be ranked, but also employee preferences for different quality levels of each characteristic are obtained this way. In a stated choice experiment (e.g. Louviere, Hensher, & Swait, 2000) different hypothetical choice alternatives are created based on an experimental design and presented to respondents in several choice sets. The respondents are asked per set to choose the best alternative based on the descriptions of the levels of the included attributes (see Figure 1 for an example of a choice set used in this study). Specifically, for each pair of hypothetical workspaces respondents are asked to choose in which workspace they could work as productive as possible for the three different work activities: individual concentration work, informal interactions, and formal interactions. In case the employees did not like any of the workspaces for a work activity, they could select the ‘neither’ option. The stated choice approach is proven to be effective in a variety of studies and research fields (e.g. Aksenov, Kemperman, & Arentze, 2016) and another advantage is that it is possible to study preferences for not yet existing or not available alternatives in real offices.

Figure 1. Example of a choice set in the questionnaire.

The hypothetical workspace profiles are created with three possible levels for each included characteristic based on an experimental design. Because knowledge workers, the respondents in this study, may have little knowledge of some technical terms, jargon should be avoided (Kamaruzzaman et al., 2018) and so simple descriptions are used that everybody would understand.

Only six of the workspace characteristics were selected to be included in the experiment. This is a common number of different variables in stated choice studies and seems to work well (Hensher et al., 2015). For the psychosocial characteristics, noise, personal control and level of enclosure are included. Privacy itself is too intertwined with level of enclosure to make it possible to create consistent, understandable workspace profiles with these two variables at opposite levels (e.g. a workspace in an enclosed environment without providing privacy). Both temperature and lighting were chosen to represent IEQ. Air quality, humidity and volatile organic compounds were not included in the experiment, because these are very hard to imagine in different levels for end-users. Ergonomics was selected to represent the individual space quality category. No further characteristics were included, as these are less strongly proven to have an effect, and more different characteristics would make the choice alternatives too complicated for the respondents to make the trade-off decision. See Table 1 for an overview of the workspace characteristics and their levels used in the experiment.

Table 1. Operationalization of workspace characteristics.

Attribute	Levels
Level of enclosure	Enclosed Semi-enclosed Open
Personal control	Full control over workspace Limited control over workspace No control over workspace
Noise	High noise levels Neutral situation Low noise levels
Ergonomics of the workspace	Specialized ergonomic furniture Ergonomic furniture Non-ergonomic furniture
Temperature	Slightly too cold Neutral situation Slightly too hot
Lighting	Much lighting Pleasant situation Little lighting

As each workspace characteristic has three levels, this would result in 729 possible alternatives (3^6 = 729). Therefore, an orthogonal fraction of this design consisting of 18 workspaces was constructed (see Hensher et al., 2015). The 18 hypothetical workspaces were randomly divided over nine choice sets, and each respondent thus evaluated all 18 hypothetical workspaces (meaning a complete design).

Fourteen companies within the authors’ networks agreed to spread the link to the online questionnaire amongst their employees. In total 321 employees participated (between 15 May and 2 June 2017). These companies were from many different sectors, but all occupied activity-based offices (see Table 2). Unfortunately, 45 participants did not fully complete the survey and therefore had to be excluded. Another 25 respondents were removed due to various other reasons, resulting in a dataset with 251 respondents. The dataset is compared to labour force statistics in the Central Statistics database of the Netherlands in order to validate the representation of the sample.

Table 2. Organizations in the sample.

Type of organization	Sector	Amount of employees	Response rate
Specialists	Healthcare	100	16%
Consultancy firm	Real estate	70	42.6%
Consultancy firm	Real estate	75	13.3%
Investment bank	Finance	35	14.3%
Learning institute	Education	20	60%
Consultancy firm	Real estate	65	18.5%
Bank	Finance	150	13.3%
Concert venue	Leisure	20	15%
University	Support staff	200	20%
University	Research department	40	37.5%
Four companies	Others	250	35.2%

Analytic strategy

The choice data is used to estimate a multinomial logit model that identifies respondents’ preferences for the workspace characteristics and their levels, and to estimate latent class models to identify segments (classes) of employees with similar preferences.

It is assumed that every employee has a certain preference or utility for a workspace for a specific activity. This utility can vary per workspace based on the quality levels of its characteristics. For the well-known multinomial logit model (Hensher et al., 2015), the utility for employee i for workspace j to perform an activity that supports productivity on choice occasion t is calculated as follows:$U_{ijt}={\text{β}}^{\mathrm{\prime }}{\mathbf{X}}_{ijt}+{\epsilon }_{ijt}$where X_ijt represent all attributes of the workspace with relative weights (parameters β’) to be estimated. The error term, ε_ijt, also represents unobserved heterogeneity. This equation assumes that all the estimated parameters are equal for all employees.

As variation in individual employee preferences for workplaces is to be expected, latent class models were estimated for each work activity to segment employees based on their preferences for workspace characteristics that support their productivity during this activity. This approach simultaneously groups respondents and estimates a separate set of utility parameters for each of the employee segments for all workspace characteristics that support their productivity. This was repeated for each of the three work activities. The utility function for employee i’s choice among j workspaces on choice occasion t, given that employee i belongs to a latent class or segment s, (s = 1, … , S), is expressed as:$U_{ijt}={{\text{β}}^{\mathrm{\prime }}}_s{\mathbf{X}}_{ijt}+{\epsilon }_{ijt}$where β’s is a segment-specific parameter vector. The probabilities of choice can be derived from the utility function. For each segment, the probability that employee i chooses workspace j for an activity at choice occasion t is:$P(y_{it}=j|\mathrm{s}\mathrm{e}\mathrm{g}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}=s)=\frac{\mathrm{e}\mathrm{x}\mathrm{p}\left({{\text{β}}^{\mathrm{\prime }}}_s{\mathbf{X}}_{ijt}\right)}{{\sum }_{j=1}^{J_i}\mathrm{e}\mathrm{x}\mathrm{p}\left({{\text{β}}^{\mathrm{\prime }}}_{\mathrm{s}}{\mathbf{X}}_{ijt}\right)}$

Segments are considered unobserved (latent) classes and have a certain segment probability:$P(\mathrm{s}\mathrm{e}\mathrm{g}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}=s)=Q_{is}=\frac{\mathrm{e}\mathrm{x}\mathrm{p}\left({{\theta }^{\mathrm{\prime }}}_s{\mathbf{Z}}_i\right)}{{\sum }_{s=1}^S\mathrm{e}\mathrm{x}\mathrm{p}\left({{\theta }^{\mathrm{\prime }}}_s{\mathbf{Z}}_i\right)}$where Z_i is an optional set of observable individual, choice situation invariant characteristics. If no such characteristics are included in the estimation, the class-specific probabilities are a set of fixed constants, which sum to one.

Then, each employee is assigned to the latent class with the highest probability. The latent class parameters can be estimated using maximum likelihood estimation (see Greene, 2001 for details). McFadden's Rho-square (Rho² = 1–LLB/LL0) is used to indicate the goodness of fit of the estimated models. To select the optimal number of segments, the minimum Akaike Information Criterion (AIC = −2(LLB-P)) is used, based on Gupta and Chintagunta (1994). To determine whether the identified segments differ regarding the personal characteristics (see Table 3 for the included characteristics), Chi-Square and One-Way ANOVA tests are executed.

Table 3. Personal characteristics included in the survey.

Personal characteristics	Source
Age (open question)	Haynes et al. (2017)
Gender (male/female)	Lai and Yik (2007)
Eduational level (primary, secondary, vocational, undergraduate, postgraduate)	Haynes et al. (2017)
Job rank (intern, regular employee, support staff, manager, board manager, other)	Hartog, Weijs-Perrée, and Appel-Meulenbroek (2018)
Working hours per week (open question)	De Been et al. (2016)
Current type of workspace (individual cell office, shared room, open 4–9 employees, open 10+ employees, free to choose)	Budie et al. (2018)

Results

Sample description

A very large proportion of the sample was highly educated (85%), which makes sense when studying knowledge workers. Regarding job rank, 14% were managers, 55% regular employees, 25% support staff and 5% interns. The current type of workspace showed that although the data were gathered in ABOs, only 27% was completely free to choose. Most worked in shared rooms (40%) or even individual cell offices (10%), with the rest working in open areas (11% with 4–9 employees, 12% with 10+ employees). The sample included 115 males (46%) and 136 females (54%), which shows only a bit more females than the general Dutch national labour force (resp. 54% and 46%, see CBS, 2017). Regarding age (M = 40, SD = 12.4) the sample showed a significantly higher proportion of people aged 26–35 than the general labour force (X²(251, 4) = 22.01, p = .000). The average working hours per week were 36 (SD = 8.5), which is also significantly higher than the national labour force (X²(251, 2) = 43.54, p = .000).

Multinomial logit models

The final dataset has 2259 records (9 choices × 251 respondents). After recoding and using an effect coding scheme (1, −1) for the attribute levels, the models are estimated. The utility values of the multinomial logit models for each activity are presented in Table 4. The significance level of each attribute level parameter shows whether this attribute level significantly influences the choice for a workspace, which serves the first part of the aim of this study. Also, for each activity category the constant is presented, indicating the preference of choosing one of the hypothetical workspaces in the choice set over the ‘no choice option’ (‘I would not choose any of the workspaces if these were the only ones to choose from’). Table 4 also shows model statistics for each of the three activity category models. As can be seen, the Rho-squares are, respectively, 0.17, 0.13, and 0.14 indicating that the model fit is acceptable for each activity category (Note that a Rho-square value between 0.2 and 0.4 is considered to be indicative of an extremely good model fit; Louviere et al., 2000, p. 54).

Table 4. Multinomial logit models results and statistics.

Attribute	Level	Activity categories
		Individual work	Informal	Formal
Constant		0.146***	1.053***	0.318***
Level of enclosure	Enclosed	0.588***	−0.497***	0.415***
	Semi-enclosed	−0.038	0.307***	−0.050
	Open (base)	−0.550	0.190	−0.365
Personal control	Full control over workspace	0.114**	0.073	0.067***
	Limited control over workspace	−0.083	0.054	−0.080***
	No control over workspace (base)	−0.031	−0.127	0.013
Noise	High noise levels	−1.034***	−0.203***	−0.872***
	Neutral situation	0.326***	0.203***	0.435***
	Low noise levels (base)	0.708	−0.001	0.437
Ergonomics of the workspace	Specialized ergonomic furniture	0.169***	0.111**	0.076
	Ergonomic furniture	0.121**	0.052	0.071
	Non-ergonomic furniture (base)	−0.289	−0.163	−0.147
Temperature	Slightly too cold	−0.215***	−0.106**	−0.219***
	Neutral situation	0.213***	0.201***	0.295***
	Slightly too hot (base)	0.002	−0.095	−0.076
Lighting	Much lighting	0.146***	0.060	0.175***
	Pleasant situation	0.300***	0.247***	0.275***
	Little lighting (base)	−0.446	−0.307	−0.450
Statistics
Log-likelihood of the estimated parameters		−2052.91684	−2153.34376	−2132.85068
Log-likelihood of the zero model		−2481.76516	−2481.76516	−2481.76516
McFadden’s Rho-square		0.173	0.132	0.141
Number of observations		2259	2256	2256
Akaike Information Criterion		1.829	1.918	1.900

*p < .10.

**p < .05.

***p < .01.

It is clearly visible that there are quite some differences in the model output between the activity categories. However, for all activity categories, the level of enclosure, noise, temperature and lighting has a significant influence on knowledge workers when choosing a workspace. For informal interactions, the attribute personal control over workspace is not relevant and for formal interactions the ergonomics of the workspace is not significant. The main difference between the activity categories is the relative impact of each characteristic on workspace choice. In order to make such differences clear, all the utility values for all attributes for the three work activities are graphically presented in Figure 2. As can be seen, the preferences per activity category mainly differ for the levels of the characteristics enclosure, personal control and noise, so only for the psychosocial characteristics. For the ergonomics attribute and for the IEQ attributes temperature and lighting, the distribution and relative impact is similar for all activity categories.

Figure 2. Utility values for the attribute levels per activity.

For individual concentration work and formal interactions, the most preferred condition is working in an enclosed environment. On the contrary, for informal interactions this is the least favoured level. Furthermore, respondents clearly indicate that high noise levels are unappealing for all activity categories, but for individual concentration work noise has a much stronger influence on the workspace choice than for informal interactions. Neither a too cold nor too hot environment is preferred for any of the activities, but a slightly too hot environment is favoured over a slightly too cold one. Little lighting is unfavourable during any of the activity categories. More appealing is much lighting or even better a pleasant lighting situation.

Based on the utility values only, it is quite difficult to determine which of the assessed variables is most important for each of the activity categories. Therefore, the relative impact of each attribute on the overall workspace preference is determined as well (see Figure 3). The relative importance is defined by considering, per activity, the range between the lowest and highest utility value of an attribute as a percentage of the overall sum of attribute ranges. The perceived most important workspace aspect during individual concentration work is noise, followed by level of enclosure, lighting, ergonomics, temperature and personal control. For formal interaction, a similar distribution was found. Again the most impactful characteristics are noise and level of enclosure, followed by lighting. For informal interactions, the order of preference ranking of the attributes is different than for the other activity categories, as level of enclosure is the most important (followed by lighting), while noise is a lot less important. Again, personal control is relatively the least important.

Figure 3. Relative impact of attributes on overall workspace preference per activity.

Latent class models

Latent class models were estimated for each different activity category as well to study the second part of the aim of this study, see Tables 5–7. The parameters for each attribute level are presented, as well as the model statistics. The optimal number of segments for each of the estimated models is based on the Rho-square (the higher the better) and the minimum AIC (Gupta & Chintagunta, 1994). The Rho-square values are respectively, 0.30, 0.18, and 0.25, indicating a good model fit and providing evidence that there are segments of knowledge workers that have different preferences for workspace characteristics when they need to perform a specific activity.

Table 5. Latent class model utility values individual concentration work.

Attribute	Level	Latent classes
		Segment 1	Segment 2	Segment 3
Constant		2.738***	−1.402***	1.044***
Level of enclosure	Enclosed	−0.137	1.042***	1.232***
	Semi-enclosed	0.130	0.018	−0.004
	Open (base)	0.007	−1.060	−1.228
Personal control	Full control over workspace	−0.023	0.246**	0.320**
	Limited control over workspace	−0.075	−0.006	−0.165
	No control over workspace (base)	0.098	−0.240	−0.155
Noise	High noise levels	−0.468***	−1.309***	−2.191***
	Neutral situation	0.136	0.497***	0.726***
	Low noise levels (base)	0.332	0.812	1.464
Ergonomics of the workspace	Specialized ergonomic furniture	0.281***	0.186	0.390***
	Ergonomic furniture	−0.009	0.270*	−0.069
	Non-ergonomic furniture (base)	−0.271	−0.456	−0.321
Temperature	Slightly too cold	−0.579***	−0.028	−0.019
	Neutral situation	0.366***	0.094	0.123
	Slightly too hot (base)	0.213	−0.066	−0.104
Lighting	Much lighting	0.236**	0.257**	0.183
	Pleasant situation	0.299**	0.566***	0.090
	Little lighting (base)	−0.535	−0.823	−0.274
Statistics Log-likelihood of the estimated parameters Log-likelihood of the zero model McFadden’s Rho-square Number of observations Akaike Information Criterion		−1750.42550 −2481.76516 0.295 2259 1.586

*p < .10.

**p < .05.

***p < .01.

Table 6. Latent class model utility values informal interactions.

Attribute	Level	Latent classes
		Segment 1	Segment 2
Constant		2.378***	−0.374**
Level of enclosure	Enclosed	−0.657***	−0.157
	Semi-enclosed	0.395***	0.128
	Open (base)	0.262	0.028
Personal control	Full control over workspace	0.072	0.069
	Limited control over workspace	0.043	0.076
	No control over workspace (base)	−0.116	−0.145
Noise	High noise levels	−0.179***	−0.332***
	Neutral situation	0.218***	0.226**
	Low noise levels (base)	−0.038	0.106
Ergonomics of the workspace	Specialized ergonomic furniture	0.136**	0.049
	Ergonomic furniture	−0.028	0.275***
	Non-ergonomic furniture (base)	−0.108	−0.324
Temperature	Slightly too cold	−0.065	−0.212*
	Neutral situation	0.161***	0.275***
	Slightly too hot (base)	−0.095	−0.063
Lighting	Much lighting	0.023	0.151
	Pleasant situation	0.253***	0.330***
	Little lighting (base)	−0.276	−0.481
Statistics Log-likelihood of the estimated parameters Log-likelihood of the zero model McFadden’s Rho-square Number of observations Akaike Information Criterion		−2025.91033 −2481.76516 0.184 2259 1.818

*p < .10.

**p < .05.

***p < .01.

Table 7. Latent class model utility values formal interactions.

Attribute	Level	Latent classes
		Segment 1	Segment 2	Segment 3
Constant		8.151	0.277	−2.442***
Level of enclosure	Enclosed	0.401***	0.236*	1.835***
	Semi-enclosed	−0.093	0.122	−0.037
	Open (base)	−0.308	−0.357	−1.798
Personal control	Full control over workspace	−0.088	0.197**	0.500*
	Limited control over workspace	−0.152*	−0.075	−0.339
	No control over workspace (base)	0.240	−0.122	−0.161
Noise	High noise levels	−0.810***	−0.936***	−2.205***
	Neutral situation	0.373***	0.487***	1.194***
	Low noise levels (base)	0.437	0.449	1.012
Ergonomics of the workspace	Specialized ergonomic furniture	0.174*	0.161*	0.401
	Ergonomic furniture	−0.031	0.128	−0.646
	Non-ergonomic furniture (base)	−0.143	−0.289	0.245
Temperature	Slightly too cold	−0.228***	−0.193**	−0.428*
	Neutral situation	0.353***	0.248***	0.208
	Slightly too hot (base)	−0.126	−0.055	0.221
Lighting	Much lighting	0.278***	0.078	0.996***
	Pleasant situation	0.192	0.393***	0.115
	Little lighting (base)	−0.470	−0.470	−1.111
Statistics Log-likelihood of the estimated parameters Log-likelihood of the zero model McFadden’s Rho-square Number of observations Akaike Information Criterion		−1865.43152 −2481.76516 0.248 2259 1.688

*p < .10.

**p < .05.

***p < .01.

Individual concentration work

Table 5 presents the results of a three-segment model for individual concentration work, with respectively 61 (24%), 100 (40%) and 90 (35%) of the respondents. The segments were tested on their relationships with the personal characteristics, but unexpectedly no significant findings came forward. Respondents belonging to segment 1 are more likely to choose one of the offered hypothetical workspaces in the experiment (constant = 2.738) and are therefore hereinafter called the ‘non-critics’. The second segment is unlikely to accept a workspace (constant = −1.402) and are therefore renamed to ‘grumblers’. The respondents in the third segment are somewhere in between (constant = 1.044) and are thus called ‘midway employees’. Besides the utility values, Figure 4 shows the relative importance of the workspace characteristics per segment.

Figure 4. Impact on overall workspace preference latent segments for individual concentration work.

Especially grumblers and midway employees indicate that level of enclosure has a strong impact on their workspace choice (utility value > 1.00 for both classes), while for the non-critics it is not significant in determining their choice. Personal control shows lower utility values for all segments, however again it is not significant for the non-critics, while the others would prefer full control over the workspace if all else equal. High noise levels are significantly unappealing for all segments, although again the non-critics show smaller utility values. Interestingly, the non-critics are the only segment that shows a significant impact of temperature, which should not be slightly too cold for them. Temperature is the most important characteristic for this segment, followed by lighting and noise. For the other two segments, temperature is the least important characteristic and noise and enclosure the most important ones.

Informal interactions

For the activity category informal interactions, a two-segment model was found to fit the data best, with 182 respondents (73%) belonging to the first segment and 69 (27%) to the second segment (see Table 6). Again, the constant values (segment 1 = 2.378, segment 2 = −0.374) imply that the respondents in the first segment are more likely to choose a workspace than the people in the second segment. Independent t-tests showed that age (p = .009) and highest completed level of education (p = .010) are significantly different between the two segments. In segment 1 the average age is 39, while in segment 2 it is 43 years. Also, in segment 1 more respondents are highly educated than in segmented 2 (with a difference of 19%). Although the age difference is not that big, class 1 is renamed ‘young academics’ and class 2 ‘aged critics’, because the latter was less inclined to choose a workspace from the offered alternatives. Figure 5 presents an overview of each workspace characteristic and its corresponding impact on the overall workspace preference per segment.

Figure 5. Impact on overall workspace preference latent segments for informal interactions.

The results illustrate that for both segments personal control is not significant. For the aged critics, level of enclosure is not significant either, while for the young academics it is by far the most important characteristic. The young academics highly dislike a fully enclosed environment, and would prefer a semi-enclosed and otherwise open environment. The aged critics care more about the IEQ (lighting and temperature), ergonomics and noise levels. Lighting and noise are also the second and third most important for the young academics. Both segments want a pleasant lighting condition and neutral temperature and noise levels. Aged critics are satisfied with regular ergonomic furniture, while the young academics prefer specialized ergonomic furniture. The aged critics significantly dislike it when it is slightly too cold and/or they have to deal with high noise levels.

Formal interactions

For formal interactions, a three-segment model was found (see Table 7), with 99 respondents (39%) in the first segment, 94 (38%) in segment 2, and 58 (23%) in segment 3. Segment 1 is very likely to choose a workspace (constant = 8.151), segment 2 is somewhat in the middle (constant = 0.277) and segment 3 is unlikely to accept a workspace (constant = −2.442). A One-Way ANOVA test reveals that only age significantly differs between the segments (F(2, 248) = 6.718, p = .001). The Tukey test reveals that the significance between segment 1 and 2 for this attribute is 0.012 and between segment 1 and 3 the significance is 0.003. Although again the average age differences are not that big, segment 1 does have the youngest respondents (M = 36.7, SD = 12) and is thus renamed ‘young non-critics’, while segment 2 (M = 41.8, SD = 13) and segment 3 (M = 41, SD = 12) are called ‘midway employees’ and ‘aged critics’ again. Figure 6 shows the relative importance of the characteristics.

Figure 6. Impact on overall workspace preference latent segments for formal interactions.

The most notable difference is that level of enclosure has a strong impact on workspace choice for aged critics in comparison to the other two segments, although it is significant for all. Aged critics strongly favour the enclosed workspace and do not like the open environment at all, while the other segments show the same preference, but it is less important for them. The aged critics care the least about temperature for formal interactions. These findings are completely opposite to the relevance of both characteristics for informal interactions. Second and third important characteristics for aged critics are noise and lighting, which are first and second for the other two segments as well. The aged critics prefer much lighting over the intermediate ‘pleasant situation’ category and strongly dislike when there is little light. Noise levels and temperature should be neutral for all segments.

Discussion

To summarize, based on this stated choice experiment we found that in general the 3 most important characteristics when choosing a workspace to support different knowledge worker activities are noise levels, level of enclosure and lighting. Mulville et al. (2016) and Lai and Yik (2007) also found noise to be more important than temperature, while Hameed and Amjad (2009, p. 6) found temperature to affect productivity the most. The latter difference might be because they conducted their research in Pakistan where the climate is different compared to the Dutch situation, but further research would be necessary to confirm this. The importance of level of enclosure confirms existing studies that this is perceived as a strong productivity enhancer (Haynes et al., 2017), just like lighting (Xuan, 2018).

As expected, the respondents preferred pleasant IEQ (light, temperature) conditions. As Faria, Inskava, and Planitzer (2017) discuss, there is no clarity about what makes a pleasant lighting, however, they showed that subjects do understand what would create discomfort. The levels for this stated choice experiment were intentionally general in wording in order to study trade-offs within and between the characteristics. A more in-depth understanding of appreciation was beyond the scope. This way the findings add to existing knowledge that IEQ is less important than the psychosocial aspects when choosing a workspace. They also show that a slightly too warm environment is preferred over a too cold environment and respondents rather have too much than little light. The preference for a too warm over a too cold workspace, might be due to the higher amount of females in the sample, as it is known that females are more sensitive to cold at work (Kim, de Dear, Candido, Zhang, & Arens, 2013) and are more often dissatisfied with climate comfort in the office (Karjalainen, 2012).

Another implication for theory is that only the psychosocial characteristics (noise, enclosure, control) show different preferences for the three specific activity categories. So where preferences for IEQ and ergonomics appear not to depend on the activity at hand, there is no one uniform solution for the psychosocial design characteristics to support all activities. Future research should shed light on whether the freedom of choice in ABO’s provides better support of psychosocial preferences than regular offices with dedicated seating. Noise, apparently, is always unappealing, but for concentrated work this is the most important issue, while for the interactive activities other things are more important. Previous studies already showed that noise is often the most unsatisfactory aspect of the office (Kim & De Dear, 2013), however, compared to other ambient factors, there is less scientific attention for and knowledge about office acoustics and noise-related behaviour. Noise can be seen as the most prevalent distraction source in offices, regardless of the office lay-out (Pejtersen, Allermann, Kristensen, & Poulson, 2006) and can lead to increased (crowding) stress for occupants (Haapakangas, Helenius, Keskinen, & Hongisto, 2008). When asked how noise affects the ability to work, only 8.4% reported that the noise in their workplace does not make their work difficult (Oseland and Hodsman, 2018). Employees who suffer from noise, are more inclined to complain about other aspects of the physical office environment as well (Haapakangas et al., 2008) and feel a loss of privacy (Bodin Danielsson & Bodin, 2009), so it is important to reduce noise where possible. This study adds to existing theory (focused on concentration related to noise) that apparently even for informal interactions, people do not prefer a noisy environment. This has also been overlooked in practice by generally offering quiet places for individual concentrated work only. For informal interactions the workspace enclosure was the most important characteristics, with semi-open (not open) as the most preferred level. Clearly, the closed layout is least preferred, which makes sense as studies have proven that this negatively influences informal interactions (Sailer & McCulloh, 2012).

Surprisingly, personal control was not very important in trade-offs with the other characteristics, while in some ranking studies this was even ranked as most important (e.g. Kamaruzzaman et al., 2018). These studies did only ask about workplace satisfaction and did not offer alternative choices, so perhaps that could explain the difference. The results did endorse the notion by for example Miller and Pogue (2009) that providing full control is better than limited or no control. Employees might not have not seen the value of personal control when choosing between hypothetical alternatives, while it could be very frustrating when you actually notice that you miss it in a real situation.

For both informal and formal interactions, age divided employees in classes with specific preferences, and for informal interactions the education level was also significantly different between the preference-based segments, confirming previous studies that age and education level influence preferences (e.g. Haynes et al., 2017; Rothe, Lindholm, Hyvönen, & Nenonen, 2011). However, the other characteristics found in existing literature (gender, job rank, working hours and current type of workspace, see Table 3) did not significantly distinguish the different segments here. Budie, Appel-Meulenbroek, Kemperman, and Weijs-Perree (2018) already stated that there is little evidence of whether and how the current workspace might influence preferences, and the findings here show no proof of this either. But for example, Al Horr et al. (2016) found that females dislike open office environments more, which could not be confirmed here. It was actually striking that especially the likelihood of choosing from the available alternatives seemed to determine the segments. Apparently, some employees are more critical than others. It would be interesting to study in future research, whether these ‘critics’ are more likely to work from home if their preferred type of workspace is not available in an ABO and they need to do individual work.

For both the interactive activity categories, the more critical segments contain (on average) older employees. It might be that age increases the likelihood of experience with more different types of workspaces, which could make people more aware of what they like and dislike, and thus more critical. For informal interactions, these ‘aged critics’ found the IEQ and ergonomics more important than their younger (and more highly educated) counterparts which were less critical in general. An explanation is lacking and should be sought in future research. For formal interactions relatively more of the ‘aged critics’ preferred a fully enclosed workspace, which cannot be explained by possible management position of more experience people, because job rank was not significantly different between the segments. Rather it might be their higher dislike of high noise levels. According to (Murphy, Bailey, Pearson, & Albert, 2018) most studies on irrelevant speech so far have failed to support the intuitive theory that older people have more problems ignoring noise due to their difficulty in keeping irrelevant information out of working memory. This study adds to existing theory that apparently they do take it into account more strongly when choosing where to work.

For one of the three segments for the individual concentrated work category, enclosure was not significant and thus not important in choosing a workspace. It is interesting to see, that apparently for some employees it is not an issue to do concentrated work in an open environment, even when it is noisy. The same segment also did not indicate to need any personal control of the workspace, thus it appears as if these employees are easy to support.

Conclusion and recommendations

In general, it can be concluded that the top 3 workspace characteristics that were found to be important when choosing a workspace to enhance activities are noise, level of enclosure, and lighting, irrespective of the activities that the knowledge worker is performing. However, different segments of knowledge workers could be identified with specific preferences, so these are not heterogeneous. Future research should identify more clearly how to determine to which specific preference-based segment a knowledge worker belongs and why.

The usage of a stated choice experiment to identify workspace preferences of knowledge workers is new in the field of corporate real estate and workplace. This study has proven its effectiveness and the results underline the importance of specific characteristics for supporting knowledge worker activities. Rather than investigating the individual effects of workspace attributes, the relative importance of six characteristics from the most important categories were studied. This approach has resulted in new insights for corporate real estate academics and practitioners.

Practical implications

Based on the stated preferences of the sampled employees, corporate real estate manager should focus first on improving noise and lighting conditions and making sure that enclosure levels meet employee preferences, in case of limited budget or time. Other characteristics were considered less important in case of trade-off decisions. Because both noise and level of enclosure are psychosocial characteristics, it could thus be worthwhile for practitioners to give these design aspects at least the same amount of attention as IEQ usually gets. There are no clear (governmental) standards/norms for enclosure and experienced noise (as long as it refers to talking colleagues or ringing phones), but this does not mean that it is less important. Spira and Feintuch (2005) calculated the cost of interruptions affecting knowledge worker productivity to be $588 billion per annum in the US alone.

Currently, in practice, a lot of attention seems to be focused on making a workspace smarter through more personal control. This research shows that in trade-offs, this is not that important, so perhaps not worth so much attention. Instead, more attention could go to the more critical and a bit older group of employees in certain organizations. If they decide to work from home to avoid the available workspaces in the office, important serendipitous informal and knowledge sharing interactions with younger staff will be impossible, hindering both mentoring and the transfer of organizational knowledge and culture/habits.

It was unexpected that personal control was not a very important aspect when choosing a workspace. Nonetheless, a lot of personal differences came forward regarding preferences, so workplace managers cannot satisfy all employees with standard conditions. The only option is then to provide employees with personal control to adjust things to their preferences.

Limitations and further research

The sample did not fully represent the Dutch national labour force, as the respondents were younger, included more females and worked more hours per week. Future studies should try to obtain a larger and fully representative sample of knowledge workers. Both gender and age have been shown to affect preferences, and age also explained the class differences that we identified. Also, the effect of working hours on workplace experience has not received much attention yet.

It is possible that people would act differently in a real-world situation than in the proposed hypothetical one. Respondents regularly indicated that none of the workspace alternatives fitted their needs and therefore chose the option ‘neither’. However, in a real-life situation it is reasonable to assume that one would still choose a workspace, since a poor workspace is still better than no workspace to work at, especially when one needs to work at the office. Therefore, complementing the results of this study with revealed data might be interesting for future studies, also in comparison with decisions to work from home.

The results do indicate that there is some variation in specific workspace preferences per employee type, but not so much as expected. Perhaps other aspects that have not been included in studies so far influence preferences more. Future studies could include additional personal, work-related and current workspace characteristics to identify whether/how these create certain preferences. For personal characteristics one could think of personality, specific (health related) needs and previous experience with other office types. For work-related characteristics, the company culture, boss–employee relationship and specific job-related activities would be interesting. More characteristics of the current workspace could be size, ergonomic quality, distance to a window, etcetera. In this study, the descriptions could intentionally be freely interpreted by the respondents, although we tried to give clear descriptions. The downside is that the respondents might not have had the same association with for example non-ergonomic furniture or low noise levels. As this is better for fair comparison, perhaps existing standards can be used in the future.

Future studies could focus on including additional workspace characteristics that might also affect workspace choice behaviour and specific employee outcomes, thereby completing the exploration of a workspace attributes ranking. If more respondents would be available in future studies, it would be possible to include more workspace design characteristics at once through different versions of the questionnaire with some of the characteristics that are randomly spread.

Filling in a stated choice experiment takes a lot of effort from participants. Future research could study whether such an experiment with pictures or perhaps using a VR-experiment makes it easier for participants to experience the alternatives and how these could contribute to their productivity. Also, studying more attributes at once in a real living-lab would be valuable for more insights. Furthermore, it would be interesting to study what the relative importance of the characteristics is for other relevant organizational and employee outcomes, such as the organization’s financial revenue. In general, using stated choice experiments more frequently in the field of corporate real estate could be beneficial to investigate the usefulness of such methods in this research area.

Disclosure statement

No potential conflict of interest was reported by the author(s).