On keys` meanings and modes: the impact of different key solutions
Transcription
On keys` meanings and modes: the impact of different key solutions
Behaviour & Information Technology, Vol. 25, No. 5, September – October 2006, 413 – 431 On keys’ meanings and modes: the impact of different key solutions on children’s efficiency using a mobile phone MARTINA ZIEFLE*, SUSANNE BAY and ALEXANDER SCHWADE Department of Psychology, RWTH Aachen University, Germany The present study investigates the impact of different key solutions of mobile phones on users’ effectiveness and efficiency using the devices. In the first experiment, 36 children (9 – 14 years) and in the second experiment 45 young adults (19 – 33 years) completed four common phone tasks twice consecutively on three simulated phones that had identical menus, but different key solutions. An approach was undertaken to quantify the complexity of keys in three models, incorporating different factors contributing to the keys’ complexity (number of key options, number of modes and number of modes with a semantically dissimilar meaning), in order to predict users’ performance decrements. As a further main factor, the degree of the users’ locus of control (LOC) was measured and interactions with performance outcomes were studied. As dependent measures, the number of inefficient keystrokes, the number of tasks solved and the processing time were determined. Results showed a significant effect of control key solutions on users’ efficiency and effectiveness for both children and young adults. Moreover, children’s LOC values significantly interacted with performance: children with low LOC values showed the lowest performance and no learnability, especially when using keys with a high complexity. From the three factors contributing to the complexity of keys, keys exerting different functions with semantically inconsistent meanings had the worst effect on performance. It is concluded that in mobile user interface design keys with semantically inconsistent meanings should be generally avoided. Keywords: Mobile phones; Key complexity; Modes; Cognitive compatibility; Usability; Menu navigation performance; Children 1. Introduction Many of the technological devices today exhibit a combination of features with crucial implications for their ease of use. They are small sized, with a diminutive display, but powerful with respect to the functionalities provided. A typical example for these devices is the mobile phone: despite these phones’ increasing number of functions the current market fulfils users’ demands for tiny devices. The difficulty of combining this aesthetic standard with the broad functionality results in complex hierarchical menus and intricate navigation key solutions differing distinctly in number of keys, their inherent functionalities, size, shape, colour and spatial arrangement (e.g. Lee and Hong 2004). Due to the demand for miniaturization and the increasing number of functions and applications, new phones need to be managed with fewer, or smaller buttons (Helle et al. 2003). From studies dealing with the effects of hierarchically structured information such as hypertext (e.g. Vicente, Hayes and Williges 1987, Kim and Hirtle 1995, Lin 2001, Pak, 2001) it is known that users have difficulties navigating through the menu, often not knowing where they are and where the sought-after function is located. However, in mobile phones this ‘cognitive friction’ (Cooper 1999) is even more distinct, mainly due to two factors. One is the small display that allows only little information to be displayed at a time and second, the ease of roaming within the menu depends on the control keys whose functioning must be transparent to the user. *Corresponding author. Email: Martina.Ziefle@psych.rwth-aachen.de Behaviour & Information Technology ISSN 0144-929X print/ISSN 1362-3001 online ª 2006 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/01449290500197086 414 M. Ziefle et al. Recent studies examining the usability of different existing mobile phones with comparable functionality showed that navigation performance is strongly affected by the different complexity of their operational logic (menu depth and width as well as key solution) (Ziefle 2002a, b, Bay and Ziefle 2005, Ziefle and Bay 2005). Mobiles with a shallow menu and only few control keys lead to better performance than complex ones. Methodologically, however, and this weakens the significance of the outcomes, it is not clear what was responsible for performance differences – the complexity of the menu, the control keys, or an interaction of both. The current study therefore addresses the impact of the complexity of control keys on the usability of mobile phones independently from the effects of the menu’s complexity, experimentally creating mobile phones with the same menu but different key solutions. In the literature, we find many guidelines regarding physical characteristics of keys and controls, such as force, displacement and sensory feedback (e.g. Sanders and McCormick 1993, Jordan 1998, Shneiderman 1998, Raskin 2000, Baumann 2001, Weiss 2002). Moreover, there are many general recommendations and design principles for the interaction of device and user (e.g. Burmester 1997, Jordan 1998, Raskin 2000, Martel and Mavrommati 2001). In contrast to the variety of information on keys and controls in general, however, to our knowledge there is no usability study dealing specifically with the differential effects of keys and their meanings in mobile phones. Moreover, there is no theoretical approach quantifying the different factors contributing to the complexity of a key solution. The total number of keys and the number of functions implemented in each key are two factors that contribute to the cognitive complexity, since users have to understand the ‘production rules’ (IF-THEN connections) of each key (e.g. Kieras and Polson 1985) and the key’s effects on the system status, furthermore, users have to learn to distinguish between the keys and remember the specific functions. Consequently, the more control keys a mobile phone provides, the more complex the key solution is because the user has to identify the target key among the remaining distracting keys at each point of the solution process. However, a small number of keys do not necessarily lead to higher usability (Raskin 2000). Namely, the convergence of many functions to single keys – in order to keep the number of keys small – also increases the complexity of keys, because the number of inherent rules for different functions within one key is usually augmented. In this context, the concept of ‘modes’ is of central importance. Modes are given if the system shows different reactions to one and the same keystroke depending on the point of the menu (Raskin 2000). An example for a well-known mode is the computer keyboard where keys lead to different reactions when the Caps Lock key is engaged. Keys that respond to modes only work reliably when the current function of the key is in the user’s locus of attention and visible or retained in the short-term memory – which is usually not the case (Raskin 2000). One could argue that the negative impact of ‘modes’ on usability and learnability of a device is nothing new and in literature concerning usability a rather frequently addressed research topic. While some researchers (e.g. Raskin 2000) recommend to generally avoid modes in a technical design (if possible) and to rely on other technical solutions instead, it is a basic question if this is also feasible in small-screen devices as e.g. mobile phones. A careful study of the different navigation keys in the different current small screen devices shows that hardly any of the devices seems to come along without modes. Presumably, the ongoing miniaturisation of the devices affects not only the screen size, which is increasingly limited, but also the size of the chassis and the space for the housing of navigation keys. In this context, the basic idea of modes seems to be an elegant escape from the shortage with respect to the amount of space. If devices are not able to house many keys (due to their restricted space), the implementation of modes and the allocation of many functions to single keys could be a proper solution. However, two points are of high ergonomic value in this context: (1) to learn if there is a sensitive cut-off between making a device as easy as possible and implementing the huge amount of functionalities that are nowadays available (Helle et al. 2003); (2) to find out if there is any combination of modes that is easier and more transparent for the user than alternative combinations, and if so, which the specific characteristics of such a combination are. This question is on the one hand rather difficult to address as the literature provides no model or approach which can predict the complexity of different key characteristics in interaction with a given menu. On the other hand this question is of high practical benefit for designers and manufacturers if answered how and how much the different factors contribute to complexity. What is possibly even more important than considering merely having modes or not, are the semantics of a key. The different functions of a key may be more or less semantically consistent, that is, similar in their meaning, which is completely disregarded by the concept of modes. The softkeys in mobile phones for example, change their function at different points of the menu according to the changing label displayed above them on the screen, e.g. ‘OK’ and ‘Select’, both meaning a confirmation action, reducing cognitive load with regard to understanding, learning and memorising the key’s meaning. If a key is once used to select an item and another time to delete, then it is rather probable to confuse the user and provokes mistakes. Also, if a key exerts its function only at one specific menu level, not working at any other point in the menu, then this On keys’ meanings and modes: children’s efficiency using a mobile phone also represents a mode; however, a semantically dissimilar mode that is definitively more difficult to comprehend and memorise than a ‘simple’ one. An experimental observation from our earlier experiments clearly shows the semantic or cognitive difficulty of such a mode: participants (young adults) had to solve a task on a mobile phone and learn to use a key (e.g. a phone book key) in one context. When they used the key at another menu level again, they noticed that the very same key that had worked perfectly one moment before now did not have any effect (as the user had moved to a menu level where the key has no function at all). Interestingly, this was a typical misinterpretation where participants did not conclude, as would have been correct, that this key has a mode with one function in a specific menu level and with no function at all other levels. They rather misunderstood the key as ‘out of order’. This differentiation between different kinds of modes leading to different degrees of cognitive demands has been completely disregarded up to now by the mode concept, instead of being treated as similarly difficult. Arguing with another theoretical concept – Norman’s (1981) activation-triggerschema system – it is assumed that action sequences are controlled by sensorimotor knowledge structures, so-called schemas. Schemas are activated and selected by a triggering mechanism that requires the appropriate condition. Through stimulus generalisation one and the same schema can be activated in similar even though not exactly corresponding conditions. This approach may help to understand why the semantic meaning of a function is of importance. If a key exerts completely different functions on different menu levels, it is not possible to develop welldefined and clear-cut schemas that are valid for single keys. Hence, performance does not reach a stage of autonomous action activation. As a consequence, the usage is therefore not easy to be learned and memorised and the identification of the appropriate key is in danger to be distracted by competing keys not at issue. 2. Questions addressed and experimental logic Since the current study is concerned with effects of navigation keys on performance, it is of central impact to examine the effects of the complexity of keys independently of the complexity of the menu. Methodologically arguing, the different key characteristics must be implemented on one and the same menu in order to isolate this factor from other confounding variables (as e.g. the breadth and width of the menu structure or shortcomings with respect to functions naming and allocation of functions to categories). Moreover, there is a number of different characteristics of navigation keys conceivable that do possibly contribute to the complexity of keys given in mobile phones. In the present study, three main key characteristics were considered as having a crucial impact on keys’ complexity, 415 even though this selection is not necessarily exhaustive, but of course arbitrarily: (1) the number of key options; (2) the number of keys having modes; and (3) the number of keys that have semantically inconsistent functionalities in different modes, representing operation rules that are more difficult to comprehend and mentally represent. Meeting demands of ecological validity, the three key features at issue were extracted from existing key models in real brands. Analysing and weighing the features differentially, a model was developed, which enables the prediction of the relative augmentation of complexity due to the three different factors present in key solutions (the term ‘solution’ is used in the ongoing text, as each navigation keys structure has its own combination of the three characteristics considered). Independent variable was the complexity of navigation key solutions in three graduations. Its effects on effectiveness and different measures of efficiency were taken as dependent variables. To get a broad insight into the effects of different key complexities in different user groups, a differential approach was pursued. In the first experiment, children were chosen to assess the specific impact of the control keys’ complexity on users who grew up with technology and who represent the users of the future. This may give good insights into the demands manufacturers and ergonomists will be facing tomorrow. It is often assumed that some usability issues will become less important because of people’s contact with technology from early on; this supposition will be questioned with this study. For a validation and a broader generalisability of the results, a replication of the study with young adults (university students) is undertaken in Experiment 2. Students can be regarded as the ‘best case’ user group, technologically prone and, rather sophisticated with respect to their cognitive capabilities, without possible limitations of the interpretation of the results due to developmental processes, neither ascending as given in children, nor descending, as in older adults. Another factor that was shown to mediate the efficiency handling technology is the locus of control regarding the usage of technological devices (Beier 1999, Bay and Ziefle 2003b, Ziefle et al. 2004). Locus of control is defined as a person’s expectation regarding the connection between their actions and their action outcomes. Internal locus of control means that a person usually attributes his success to his own competency, while people with external locus of control ascribe their action outcomes not to themselves, but rather to chance or others. When transferring the concept to the context of using technology, the degree of a person’s own confidence handling technological devices is of central interest. However, and this is not known so far, it is unclear whether the concept applies also for children’s selfestimation of their performance level. Thus it was examined whether there is a correlation between children’s and young 416 M. Ziefle et al. adults’ individual beliefs about their abilities to competently use mobile phones. Furthermore, it is very insightful to determine possible interactions between key complexity on the one hand and the degree of feeling competent with technology on the other hand. Therefore, in this study, besides processing tasks on mobile phones that differ in their navigation key solutions, the participants completed a short version of Beier’s (1999) questionnaire assessing the locus of control regarding technology. 3. Pre-study In a pre-study it was assessed which control key solutions and which menu may be used. In this context it was a general question if a complex menu necessarily needs to be operated with a complex key solution, as this is often the case for real devices. Therefore, an extensive pre-study with computer simulations was carried out, in which existing menus of real mobile phone devices with existing key solutions were emulated. To combine an experimental design (variation of one factor while holding confounding factors constant) with demands of ecological validity, three key solutions of existing models were selected and adapted in order to operate one and the same menu, also originating from a real mobile phone. The rather complex (deep) menu structure of the Siemens C35i was chosen and it was found out that only a few changes in the software of the phones’ menu were necessary to allow an operation with the key solution of two other widespread models for the mass market – the Nokia 3210 and the Siemens S45 – in addition to the original key solution. As shown in the following sections, the three key solutions differ considerably, regarding the number of keys, their inherent rules and their overall complexity. The Nokia key solution consisted of few and very simple keys, and the Siemens S45, with the same number, but less complex keys than the original keys of the Siemens C35i (cf. table 1). The fact that the complexity of keys is not necessarily connected to the complexity of the functions implemented is of importance, last but not least for methodical and operational issues of the current experiment. In order to refer the outcomes exclusively to the effects of the different key complexities, it was a central methodological requirement to ensure that in each key-menu combination the same number of steps were necessary to solve the task on the shortest way possible. This was achieved with our solution and will be described in the next sections. 4. A quantitative approach to key complexity In the following, a prediction model is developed on the basis of three factors contributing to keys’ complexity, which have not been differentiated up to now: the number of options to be pressed in a key solution, the number of different functions a key can exert (modes) and the number of semantically dissimilar modes. In the three predictions the three factors are weighed differently, thus resulting in a different complexity of the three key solutions and as a direct result in a differing predicted efficiency and effectiveness handling the mobile phones. It should be noted that any proposed model and its validity naturally depends on the sensitivity and the extent of differentiation of the factors that are implemented into the model. It is a major concern of the study that all three key features proposed have similar characteristics for the three tested mobile phones. Hence, any correspondences differ only for their proportions, which were selected as reasonable. However, the selection is arbitrary, which means that more and different parameters may equally contribute to cognitive complexity. Nevertheless, the three parameters seem to reflect very basic key characteristics of real brands and are therefore implemented into the model. On the base of the fit of predictions with the empirical findings, other characteristics, disregarded so far, will be discussed. Before the model is developed, first the single keys of each solution will be described in detail. 4.1 Key description The three solutions differed with regard to the number of keys (more specifically, the number of different options to be pressed), the number of keys with modes, the number of different functions each of these keys can exert, the number of semantically inconsistent functions as well as the number of redundant key functionalities. To describe the complexity, the key solutions were analysed according to the different functions the keys exert in the four tasks that had to be solved by participants: calling a number, sending a short message, hiding one’s own number and setting a call divert. An overview on each of the keys’ modes of operation within the three solutions is provided in table 1. According to table 1, the Nokia 3210 navigation keys (in the following referred to as Phone A or keys of Phone A) exhibit four key options and two of them are sensitive to modes. The c-key is used for corrections of letters and digits as well as for returns to higher menu levels, therefore having modes, yet these two functions can be regarded as semantically consistent, as they both mean ‘undo’. The centrally positioned key is a soft key used to enter the menu, to select highlighted menu entries, to confirm and to effect calls (four functions, three of them semantically similar representing confirmation actions, but entering the menu is not a confirmation action and can therefore be regarded as semantically dissimilar). The scrolling key effects movements up and down within any level of the menu (two functions but no mode). The Siemens S45 (in the following referred to as Phone B or keys of Phone B) has overall, eight key options. There are two soft keys used to carry out the function that is On keys’ meanings and modes: children’s efficiency using a mobile phone 417 Table 1. Description of the keys present in three different keys solutions (Nokia 3210 (phone A), Siemens S45 (phone B), Siemens c35i (phone C)). The keys are structured within the number of key options, the number of keys with modes, the number of functions per key at a time, the number of modes per key, the number of dissimilar functions and the number of redundantly allocated functions. displayed above them in the display. The left soft key has six functions (enter the phone book, access the mailbox, change, correct, save and sometimes no function at all), four of those being semantically dissimilar functions (select the phone book, save and access the mailbox, as well as ‘no function’). ‘Change’ and ‘correct’ can be regarded as semantically similar. The right soft key is used to enter the menu, to select a highlighted function, to send and to confirm (four functions, two semantically consistent confirmation actions and two dissimilar actions). Moreover, there is a big round button with four direction arrows (no modes). The arrows pointing up and down always have the function of scrolling up and down within the menu. The arrows pointing left and right are only used to select menu entries and step back to higher menu levels (in the tasks carried out here). These are redundant functions because the buttons with arrows pointing to the left and right have the same functionality as the right soft key (when used to select) and the red receiver button for returns to higher menu levels. These redundant functionalities may decrease the complexity of the solution because actions can be carried out in two ways (Sanders and McCormick 1993). Furthermore, there is a green receiver key on the left side to make calls when digits have been entered (one semantically dissimilar mode, as it works only on very few occasions) and a red receiver button on the right side to end calls as well as returning to higher levels in the menu (two modes, one being semantically dissimilar from the other). 418 M. Ziefle et al. The C35i solution (in the following referred to as Phone C or keys of Phone C) consists in total of seven key options. Each of the two rocker switches contains two options (marked by a dot on each of side of the rocker switch) were the key to be pressed, thus resulting in two ‘stroke options’ per key. Sometimes (depending on the menu level) however, there are not two different options to be selected, but the same function is exerted, independently of the side of the rocker switch that was pressed. This is shown by the label on the display above the specific key or part of the key, respectively. The left rocker switch has six different modes. The functions exerted are scrolling (left: up, right: down), selecting the mailbox, changing, saving entries, and sometimes it has no function at all, depending on the point of the menu. Four actions are semantically very different from the scrolling function. The right rocker switch serves to enter the menu, to select, to correct (left part) and confirm (right part), or to correct (left) and save (right), and to send a message (eight functions/combinations of functions, where six are semantically dissimilar from selecting/confirming). Additionally, there is an extra key with an icon (open book) to enter the phone directory, which is not active most of the time, thus representing two modes, one being semantically dissimilar from the other. Furthermore, a big, centrally positioned key with a green receiver is used to make calls, also only exerting a function at specific points of the menu, otherwise having no function (one function being semantically dissimilar from the second). Finally, Phone C has a smaller key with a red receiver sign to end calls as well as for hierarchical steps back in the menu (two modes where one is a semantically dissimilar function). 4.2 The prediction of cognitive complexity In order to estimate how complex a certain key solution – the combination of all control keys – is for the users, different parameters can be taken into consideration. 4.2.1 Prediction on the basis of the number of key options (all keys treated as equivalent). As users have to make a decision at each point of the solution process concerning which of the different key options is to be pressed, firstly the number of possible key choices should be regarded. The resulting complexity of the three key solutions for each of the four tasks considering only the key options is calculated in the second line of table 2. Phone B, for example, consists of five keys but one of them has four options, thus the complexity would be eight. The table shows that with this parameter the lowest complexity is found for the keys of phone A. Compared with this solution phone C is 75% more complex and the keys of phone B are the most complex with 100% higher complexity than phone A (cf. table 2). 4.2.2 Prediction on the basis of the number of key options, modes and redundant functions (all keys and modes were treated as equivalent). A second factor contributing to complexity – as mentioned before – is the number of different modes each of the keys has, that is, the number of functions that may be carried out with each of the keys. In this model, the number of options per key was multiplied with the number of modes it has and added. As having redundant functions implemented should ease the use of the keys, the number of redundant functions was subtracted from the total. This leads to the predicted complexity found in line three of table 2. Taking again phone B as an example, it is shown in table 1 that four of its five keys have modes, thus exerting more than one function. Adding the number of functions each option can exert (6 þ 4 þ (4 6 1) þ 2 þ 2) and subtracting the two redundant functions, a complexity of 16 is calculated for this key solution. With these parameters again the keys of phone A result in being the least complex solution and phone B is again 100% more complex than phone A. Phone C, though, is now the one with the highest complexity, namely 150% higher than phone A and 25% higher than the key solution of phone B (cf. table 2). 4.2.3 Prediction on the basis of the number of key options, modes and meanings (semantically dissimilar modes weighed by factor 2) as well as redundant functions. The third relevant factor mentioned in the Introduction section is the fact that functions exerted by one and the same key can be semantically similar or dissimilar. As (semantically) dissimilar modes are supposed to increase users’ confusion about a key, this is introduced in the next model predicting cognitive complexity by weighing each dissimilar mode with the factor 2, that is, they contribute with double the strength of a semantically similar mode to cognitive complexity. Again, keys and modes are multiplied and the two redundant functions are subtracted (for phone B). For phone C keys this means that when adding the total number of dissimilar functions (in total, eight) to the complexity calculated in the former model (in total, 16) we arrive at a complexity of 24. Thus, this last model leads to the prediction that compared to the key solution of phone A, phone B is 167% more complex and phone C is even 244% more complex. The increase in complexity from phone B to phone C is 29%. The predicted differences in cognitive complexity between the three key solutions should lead to comparable differences in users’ performance when interacting with the keys. Figure 1 visualises the cognitive complexity of the key solutions predicted with the three models (key options only, key options plus modes or key options, modes plus meanings). If semantically dissimilar modes do indeed have the dramatic impact on usability that we assume, performance outcomes should best match with the predictions of model 3. 419 On keys’ meanings and modes: children’s efficiency using a mobile phone Table 2. Predicted increase in complexity on the basis of different model assumptions (Model 1: key options; Model 2: key options6the number of modes – redundant options; Model 3: key options6the number of modes6number of dissimilar modes – redundant options) for three key solutions (Nokia 3210 (phone A), Siemens S45 (phone B), Siemens c35i (phone C)). Phone A Phone B Phone C Model 1: Key options 4 8 þ100% 7 þ75% Model 2: Key options6number of modes – redundant options 8 16 þ100% 20 þ150% Model 3: Key options6number of modes6number of dissimilar function – redundant options 9 24 þ167% 31 þ244% Figure 1. Predictions with respect to the impact of cognitive complexity in three models: white bars (model 1): predictions on the basis of key options; grey bars: predictions on the basis of key options and modes (model 2); black bars: predictions on the basis of key options, modes and semantically dissimilar modes. 5. General method In this section, the methodological details are given that refer to both experiments. This includes the operationalisation of the independent and dependent variables, the material used as well as the technical hardware and the software design. Specific aspects concerning one of the two experiments will then be described in the respective sections. 5.1 Variables The first independent variable was the complexity of the navigation key solution of the mobile phones in three steps (as defined in table 2, phone A, B and C). The menu of the phones was identical representing the original Siemens C35i menu, which in previous studies (Ziefle 2002a, b, Bay and Ziefle 2003a, 2005, Ziefle and Bay 2005) had shown to be rather difficult to use. The second independent variable is the locus of control regarding the use of technology. Therefore users were divided into two groups by median split. As dependent variables of the experiment, three different measures were focused: First, as the most direct measure, the number of inefficiently used keys were analysed. These were all keys pressed while completing the tasks, which did not lead to any or no task-related reaction on the display. The following keystrokes were specified: soft keys, arrow keys, green receiver keys within the menu, red receiver keys, c-key or the phone book key when not exerting a function, hash and asterisk at any point in the menu, digits except when task relevant. The different inefficiently used keys were summed up to the total number of ineffective keystrokes. Second, as a measure of effectiveness, the number of successfully solved tasks was determined. Third, as efficiency measure, the time on task was assessed. 5.2 Locus of control interacting with technology The test assessing users’ locus of control regarding the use of technology (LOC) was developed by Beier (1999). It is available in a long version consisting of 24 items and a short version with eight items. According to Beier’s own studies the reliability of the short version shows a Cronbach’s Alpha of .89. A second study with 36 participants yielded a reliability of 0.90 (Beier 1999). For the first study presented here with children as participants, the original items were adapted to a language that is appropriate for them. Since ‘technical problems’ are of central interest, it was ensured that the children knew exactly what was meant. The experimenter defined ‘Technical problems’ with examples from the children’s everyday life, such as to tape-record a movie on the videocassette recorder, to handle video games, to repair the gear change of a bike or working on the PC. The children had to answer eight questions measuring their LOC regarding the usage of technical devices. As examples for the items used in this questionnaire, children were asked 420 M. Ziefle et al. whether they think that many technical devices are too difficult to understand and use or whether they enjoy cracking a technical problem if present. The participants’ task was to confirm or deny on a six-point-scale (from ‘not at all true’ to ‘absolutely true’) the respective questions. 5.3 Tasks Four very frequent telephone tasks were selected. In order to see whether complex solutions can be learned easily, a second trial was run, where the tasks had to be completed in a slightly modified manner: 1. Calling a number: users had to enter 10 digits (the phone number) and press one control key. 2. Sending a short text message (in order to control the differences in the speed of typing, the message was already provided and only had to be sent when participants reached the adequate point in the menu): Users had to perform 11 steps within the menu and enter 10 digits (the phone number where the text message had to be sent to). 3. Hiding one’s own number when calling somebody: 19 steps had to be executed. 4. Making a call divert: 14 steps were necessary. 5.4 Material The three mobile phones were simulated as a software solution and run on a PC, which was connected to a touch screen. The original key solutions differed in a number of factors apart from the focused functions exerted in this study. To interpret solely the keys’ meanings and modes, ruling out any other key characteristics or phone features to be possibly confounding (size, haptics, the interspace between keys, etc.), the simulated phones were carefully matched in their appearance (same proportions in physical dimensions such as keys’ size and inter space as well as display, font type and letter size). In order to avoid any biases, a virtual mobile phone was presented on the screen, not resembling any specific brand, but rather kept ‘neutral’ and unobtrusive without eye-catching colouring or prominent features. It was intended that the emulated mobile phones were not in the foreground, but rather the operational logic in the three key solutions. As children may have problems with the not yet fully developed psychomotor abilities that are necessary to hit the phone buttons properly, buttons and text layout were somewhat increased, ensuring proper handling and good viewing conditions. Providing optimal information density on the display (Bay and Ziefle 2004), three menu functions could be seen on the display at a time. Users’ actions were logged online in order to reconstruct in detail which key was pressed and how often and at which point within the menu. It should be noted that issues of tactile feedback of the keys were not considered; therefore, the simulations on the touch-screen should provide valid results. 6. Experiment 1: children interacting with keys differing in complexity In the following section, an experiment conducted with children using mobile phones with three key solutions differing in complexity while holding the menu constant will be reported. 6.1 Participants Thirty-six children between 9 and 14 years (M ¼ 11 years) volunteered to take part in the study. The children had answered to announcements published in different schools by a letter addressed to children and their parents in which they were invited to join an experimental study dealing with the usability of different mobile phones. As the usage of technology and, especially, mobile phones is highly estimated in child users, children responded highly motivated to join the study. The 36 children were matched by age and gender to the three experimental groups in such a way that homogenous groups resulted. In each of the groups, five boys and seven girls participated, with a mean age of 11.3 years in the first group (phone A), a mean age of 10.8 in the second experimental group (phone B) and a mean age of 10.8 in the third group (phone C). Moreover, the questionnaire assessing the LOC when using technological devices was carried out before the experiment. The three experimental groups did not differ significantly with regard to this cognitive style (phone A: M ¼ 64.7 points; phone B: M ¼ 68.9 points; phone C: M ¼ 68.5 points, out of a maximum of 100 possible points). To ensure that differences in navigation performance are due to the different complexities of keys and not to different experience with other technical devices, a very detailed interview was undertaken before the experiment assessing the children’s expertise. . Frequency of usage. Participants were to state if and how often they use technological products (mobile and wireless phone, Fax, PC and video cassette recorder (VCR)), using a 5-point scale (1 ¼ several times per day, 2 ¼ once per day, 3 ¼ once or twice a week, 4 ¼ once or twice per month and 5 ¼ less than once or twice a month). . Interest in technology. Further, the children rated their general interest in technology, using a 4-point scale (1 ¼ very low interest, 2 ¼ low interest, 3 ¼ high interest, 4 ¼ very high interest). 421 On keys’ meanings and modes: children’s efficiency using a mobile phone . Ease of use. Moreover, the estimated ease of use when using different technical devices had to be stated again on a scale with four answering modes (1 ¼ the usage is easy, 2 ¼ the usage is quite easy, 3 ¼ the usage is quite difficult and 4 ¼ the usage is difficult). The outcomes of the self-reports of the children shows a picture of a user group that is rather experienced with technology (table 3). As can be seen in table 3, the children use a wireless phone daily, but, however, the mobile phone only twice a week. Twenty-three of the 36 children reported using it mostly only to make and answer calls; 13 children use the mobile also to send text messages or for more complex functions. The children started to use the PC daily, and a VCR about once per month, while the fax machine is used less than once per month. The overall interest in technology was medium (M ¼ 2.1 points). With respect to the selfreported ease of use of technical devices, the handling of a wireless fixed-line phone is among all other devices rated as most easy. It was judged that the PC and the mobile phone as well as the VCR were handled quite easily. Only the handling of a fax machine is reported to impose a more difficult usage. With respect to interest in technology in general, the children reported having a moderate interest in technology in all three groups (phone A: M ¼ 2.1, phone B: M ¼ 2.0 and phone C: M ¼ 2.3 points), not differing from each other, as ensured by non-parametric Kruskal-Wallis Tests. Except the difference between groups (the frequency of using a wireless fixed-line phone was slightly more frequently used in the phone B group), no significant differences with respect to the previous experience with, and interest in, technology in general and the judged ease of use were detected between the three experimental groups. Thus, differences in the performance can be interpreted as a function of the navigational keys’ complexity and confounding effects of the children’s differential previous experience can be ruled out. 6.2 Procedure At the beginning of the experiment, the users’ previous experience with technological devices was assessed. In order to get familiarised with the experimental apparatus, especially the handling of the touch-screen, this questionnaire had to be completed using the touch-screen. Then, the experimenter assessed the children’s locus of control regarding the use of technological devices by reading out the eight items, which had to be answered by the children on the 6-point scale. Afterwards, the children completed four telephone tasks that were applied twice in a slightly modified manner in order to measure effects of learnability. A time limit of 5 minutes was set for each task. It was ensured by usage of a child-friendly language that the children understood exactly what they had to do in each task. They were instructed to solve the tasks as fast and thoroughly as possible. If a task was solved successfully, a ‘Congratulations’-message appeared on the display. User manuals were not provided. After the experiment, the children were gratified for their efforts with a small present they could choose from a variety of toys. Depending on the individual working speed, the whole experiment for one participant lasted between 30 and 50 minutes. 6.3 Results The results were analysed by multivariate analyses of variance and when assessing task difficulty, with analyses for repeated measurements. The main factors were the navigation key solution and the degree of LOC (dichotomized by median split into a group with higher and a group with lower values). The dependent variables were the number of inefficient keystrokes in completing the four tasks, the effectiveness (number of correctly solved tasks) and the time needed to process the tasks. For determining the overall complexity of the key solution, the performance in the eight tasks was summed up. Assessing learnability effects, the first trial of the four different tasks was summed up for each measure and contrasted to the performance in the second trial. Finally, correlations of different user characteristics (previous experience with technology, locus of control) and navigation performance were undertaken. 6.3.1 Effects of key solution. In order to determine the overall effect of the key complexity, a MANOVA was Table 3. Mean ratings of children’s self-reports with respect to the experience and judged ease of using technology. Mobile phone Wireless phone Fax PC VCR The frequency of using a device (1 ¼ several times per day, 2 ¼ once per day, 3 ¼ once or twice a week, 4 ¼ once or twice per month and 5 ¼ less than once/twice a month) The ease of using a device (1 ¼ the usage is easy, 2 ¼ the usage is quite easy, 3 ¼ the usage is quite difficult and 4 ¼ the usage is difficult) 3.5 2 4.9 2.4 3.9 1.77 1.1 2.4 1.75 1.8 422 M. Ziefle et al. carried out with the number of inefficient keystrokes, the effectiveness (number of tasks solved successfully) and the time on task. Between subject variables were the complexity of keys (three groups: phone A key solution, phone B key solution and phone C key solution) and the degree of locus of control (two groups, high and low competence, segregated by median split). The MANOVA analysis yielded significant omnibus effects of the cognitive complexity (F (2, 52) ¼ 1.8; p 5 0.1). The degree of LOC alone did not significantly affect the performance, but the combination of both factors, the key complexity and the degree of LOC, were shown to significantly interact (F (2, 52) ¼ 2.2; p 5 0.05). The outcomes are visualised in figure 2 for the three dependent measures (left side: number of inefficient keystrokes; centre: number of tasks solved correctly; right side: time on task). In all measures, the factor complexity of key solution affects performance in the same way: it was always the key solution of phone A which showed the best performance (number of inefficient keystrokes: M ¼ 9.8; number of tasks solved: M ¼ 6.2 (out of eight); time on task: M ¼ 1074 s) and the worst performance for phone C (number of inefficient keystrokes: M ¼ 42.6; number of tasks solved: M ¼ 5.6 (out of eight); time on task: M ¼ 1220 s), with the phone B key solution ranging between these two (number of inefficient keystrokes: M ¼ 19.9; number of tasks solved: M ¼ 5.9 (out of eight); time on task: M ¼ 1087 s). With respect to the single F-tests, the number of inefficient keystrokes was significantly affected by the different key solutions (F (2, 52) ¼ 6.3; p 5 0.01), while for the number of tasks solved and the time on task, the significance level was not reached. Now, the interaction of the keys’ complexity and the degree of LOC is of interest. As the analysis showed, again, the number of inefficient keystrokes was the most sensitive measure, yielding significant effects (F (2, 52) ¼ 4.6; p 5 0.05), while the other measures did not reach significance alone but contributed to the omnibus value. Figure 3 illustrates that the pattern of results was similar for all measures, depending on the complexity of the key solution. Regarding the number of inefficient keystrokes (figure 3, left side) it can be seen that with the phone A keys, all children, independent from their degree of LOC, showed the lowest number of inefficient keystrokes (low: M ¼ 10.1 inefficient keystrokes; high: M ¼ 9.2 inefficient keystrokes). For phone B, the ‘gap’ between both LOC groups is slightly, but non-significantly bigger, with the group with the low LOC pressing the keys 24 times inefficiently and the high LOC-group, 16 times. A very dramatic difference occurred in the phone C key solution: while children with high LOC values performed rather well with, on average, 17.5 inefficient keystrokes, the low LOC group yielded the worst performance (M ¼ 92.8 inefficient keystrokes). The other two dependent variables showed a comparable picture. In the phone A solution, both LOC groups solved the same number of tasks: on average 6.2 out of eight tasks were solved successfully. Both more complex key solutions were much harder to grasp for the low LOC children, with only 5 tasks (phone B) and 4.7 tasks (phone C) solved correctly. For the children with high LOC values, the performance of all key solutions lay close together (phone B: M ¼ 6.8 tasks, phone A: M ¼ 6.2 tasks and phone C: M ¼ 6.1 tasks). Finally, looking at time on task, the pattern showed to be the very same. The phone C keys performed worst again, with the strongest effect for children with low LOC values, needing 1442 s to process the tasks with the phone B. With the phone A keys, the children needed only Figure 2. Effects of key complexity on the number of inefficiently used keys (left), the number of correctly solved tasks (centre) and the time on task (right) for the children group (N ¼ 36). On keys’ meanings and modes: children’s efficiency using a mobile phone 1070 s and the phone B was ranging between both other keys (M ¼ 1322 s). 6.3.2 Learnability. This analysis focuses on the question whether the cognitive complexity of phone keys can be understood in a second trial, that is if the children were able to grasp the rule of the keys and use them correctly in the second trial. Again, the most obvious effects were found for the number of inefficiently used keys. In figure 4, left side, the first trial is contrasted to the second trial in all three key solutions. Even if the overall number of inefficiently used keys dropped significantly (F (1, 30) ¼ 3.9; p 5 0.1), the differences between the three key solutions remained, even in a second trial (first trial: F (2, 52) ¼ 3.8; p 5 0.05); 423 second trial: (F (2, 52) ¼ 5.8; p 5 0.05)), corroborating that the key’s logic was not caught equally well in the three key solutions. It was phone C that caused the largest number of inefficient keystrokes in the first (M ¼ 22.8) as well as in the second trial (M ¼ 19.8). In contrast, the number of inefficiently used keys was 7.7 in the first trial and only 2.1 when using the phone A keys, and in phone B the number of inefficiently used keys was 14.8 in the first, and 5.1 in the second run. Thus, it can be shown that a benefit by training cannot be found in the rather complex phone C key solution, while the other two key structures were understood by the children, taken from the distinct performance increments in the second compared to the first trial. A closer look into the learnability results in both Figure 3. Effects of key complexity and degree of LOC on the number of inefficiently used keys (left), the number of correctly solved tasks (centre), and the time on task for children with low LOC (black lines) values and high LOC values (grey lines). Figure 4. Effects of key complexity and learnability on the number of inefficiently used keys for the first and the second trial (left), and for children with low and high LOC values in the first (centre) and second trial (right). 424 M. Ziefle et al. LOC-groups (figure 4, centre and right side) showed the combination of having a low LOC and using the phone C solution with many dissimilar modes to be most fatal, not only in the first (F (2, 52) ¼ 3.1; p 5 0.1), but still and even stronger in the second run (F (2, 52) ¼ 4.3; p 5 0.05). As before, the patterns regarding effectiveness and time on task were comparable, but did not reach the significance level. In the first trial, using the phone A keys, the children solved on average 3.1 tasks in the first and in the second trial, needing 587.8 s when accomplishing the tasks for the first time and only 477.1 s in the second, trial. In the phone B condition, the number of tasks solved was 2.6 in the first trial and 3.2 in the second accompanied by a processing time of 636.6 s in the first and only 442 s in the second trial. Once more, the worst performance was found in the key solution of phone C: in the first trial, the children solved, on average, 2.9 of the four tasks, thereby needing 668 s. In the second run, 2.8 out of four tasks were solved correctly in 552 s. 6.3.3 Key level. Up to this point, the inefficiently used keys in the three key solutions were globally analysed by comprising all different types of inefficiently used keys. It might be continuative though to have a look at specific key types that were used inefficiently. Thus, in this section, a very basal insight into selected types of inefficiently used keys is provided. Four types of keys were descriptively analysed: (1) examples of modes with a semantically similar meaning; (2) example of a key with one function, which is only exerted at certain points, therefore regarded as semantically dissimilar mode; (3) examples of keys with different modes that are semantically different; and (4) examples of inefficiently used keys that were at no time helpful for the processing of the tasks, but might be misunderstood as target-oriented. 1. Key with two, but semantically similar modes. As an example the ‘C’-button on the phone A is analysed (cf. table 1). The key has two functions, making corrections and going back to higher levels in menu hierarchy. The data show that the key was used 19 times inefficiently in the first trial. In the second trial, the key was not used inefficiently anymore, showing that the mode was fully understood by the children. 2. Key with two, but semantically dissimilar modes. In this analysis, a key is selected that exerts its function only at one specific level having, however, no function at all other levels. It concerns the green receiver key of both Siemens models (cf. table 1). It is centrally positioned on phone C; on phone B it is placed laterally. In the first trial, the green-receiver key is inefficiently used 34 times with phone B and still more often, namely 49 times with phone C. Does the misuse drop in the second trial? With phone B, this was the case. Compared to the first trial, the children used this key 50% less inefficiently (17 times). With phone C, no learnability was present; rather the opposite, the green-receiver key was used even more often than in the first trial (50 times). How can this result be explained? Two keys with exactly the same mode lead to a very different performance and learnability. The meaningful difference between them is supposedly their spatial position on the keypad. In contrast to phone B, the key of phone C is designed in an eye-catching way, being centrally positioned (compared to the lateral position of the key in phone B). Thus it can be assumed that its position was responsible for the frequent misuse, presumably misinterpreted by the children as having the function of ‘enter’ or ‘confirmation’. 3. Keys with many and semantically dissimilar modes. Next, mode keys exerting many functions and, especially, having dissimilar modes are of interest. In phone C, the softkeys (rocker switches) have definitively more key options in general and more semantically dissimilar functions compared to phone B. If having semantically dissimilar modes is especially cognitively demanding as proposed by model 3, then the frequency of inefficient use of these keys in phone C should clearly exceed those with phone B. The analysis showed that this was indeed the case. In the first trial, the softkeys were used 25 times more inefficiently with phone B. With phone C however, the keys were even used 92 times in the first trial. Looking at the second trial, where children should have learned the specific meaning of the keys, it was found that these keys were definitively hard to understand. Even if there was a general decrease in the usage of the keys, they were still used 16 times with phone B and 72 times with phone in the second trial. Thus, it can be concluded that keys with many and especially semantically dissimilar modes cause a high cognitive load that is still present when the children have had some time to learn the functioning of the keys. 4. As examples for keys never contributing to the solution of the task, but possibly indicating the children to run out of ideas how to proceed constructively, the frequency of using the keys ‘#’ and ‘*’ was enumerated. Both keys are in the keypad of the mobile phone, to the left and right side, flanking the ‘0’ in the lowest row. In figure 5, the frequency of using these two keys is illustrated for the first and the second trial in all three key solutions. As can be seen, the different complexities of the key solution can even be seen in the usage of these two irrelevant keys. In the first run, hash and asterisk were used eight times with phone A, and 23 times with On keys’ meanings and modes: children’s efficiency using a mobile phone Figure 5. Effects of key complexity on the number using # and * as inefficient keys in the first and the second trial. phone B. The highest number of inefficiently used keys was found for key solution of phone C (30 times). In the second run, the children generally showed a distinctly lower frequency of using these keys: with phone A, these keys were not used anymore and with phone B only three times, hinting at a learning process. With phone C however, these two keys were still used 20 times. Apparently, even if the usage of these keys should not be linked to the complexity of the key structure itself, as these keys are given in any phone, they rather reflect the difficulty the children experienced using the mobile phones, possibly in the sense of a more general searching-for-help pattern. 6.3.4 Correlations between the dependent variables. Now, for the full understanding of the outcomes and the identification of potential predictor variables, it might be insightful to look at possible interrelations between the children’s self-reported experience with technology (assessed in pre-experimental screenings) and the performance outcomes on the one hand as well as the degree of LOC on the other hand. Thus, correlations, for ordinal data Spearman-Rho and for nominal data Kendall-Tau-values, are reported. The degree of LOC was significantly correlated with all performance measures (number of inefficient keystrokes: r ¼ 70.47; p 5 0.01; number of tasks solved: r ¼ 0.5 (p 5 0.01; time on task: r ¼ 70.36; p 5 0.05), showing that children with high LOC values performed distinctly better. Having a mobile phone (r ¼ 70.31; p 5 0.01), the frequency of using it (r ¼ 70.32; p 5 0.01), and the reported ease handling a mobile phone (r ¼ 70.45; p 5 0.01) were further factors that were revealed to be significantly correlated with the degree of 425 LOC. But not only was mobile phone possession, the frequency and ease of its usage interrelated with the degree of LOC, but also the frequency of using a PC (r ¼ 70.39; p 5 0.01) and a VCR (r ¼ 70.35; p 5 0.01), as well as the reported ease of using the VCR. The reported general interest in technology was neither significantly correlated with LOC, nor with the frequency of using technical devices, as a mobile phone, a VCR, or a fax copier, even if one would have expected that. Rather, the LOC degree was correlated with the estimated ease using a PC (r ¼ 70.33; p 5 0.05). Even if the LOC values of girls were significantly lower than boys’ LOC values (r ¼ 70.38; p 5 0.01), gender showed no correlation with the number of inefficiently used keys (n.s.) and time on task (n.s.), but for the number of correctly solved tasks (r ¼ 70.28; p 5 0.1). The last finding may be interpreted such that the degree of LOC worked as a motivational regulator for task success. Thus, concluding the correlation outcomes, children’s LOC values reflect the general degree of the reported experience with technological devices and the reported ease using them. 7. Experiment 2: young adults using different key solutions in mobile phones To validate the results found in Experiment 1 with children as participants, the study was replicated with 45 university students. They can be considered highly experienced in the use of technical devices in general and should represent the optimal user. If there is an impact of the different cognitive complexity of the key solutions on the performance in this user group, a high generalisability of the results can be assumed. 7.1 Participants The age of the 45 participants ranged between 19 and 33 years (M ¼ 22.6). Thirteen were male and 22 female, distributed equally to the three experimental conditions. The LOC did not differ significantly in the three conditions (F (2.44) ¼ 1.93; n.s.) with values of 67.6 in phone A group, 76.9 in phone B condition and 76.9 in the phone C group. While 32 of the participants reported possessing a mobile phone of their own, 13 actually did not. The 13 novice users were also distributed equally to the three experimental groups. In phone A and phone C conditions there were four novices each and five novices used phone B to process the tasks. Through nonparametric Kruskal-Wallis tests it was made sure that the three groups also did not differ with regard to their experience and their judged ease of using different technical devices. The only significant difference was found with regard to the functions of their mobile phone, which the participants reported to use (w2(2) ¼ 5.3; p 5 0.1). Participants in phone C group reported on using 426 M. Ziefle et al. more sophisticated functions than just making calls and sending text messages on average (M ¼ 5), whereas phone A group reported on only using these two functions (M ¼ 3.9) and the phone B group ranged in between (M ¼ 4.6). Overall, the participants reported using their mobile phone mostly daily or once or twice a week (M ¼ 2.5) and rated the use as rather easy (M ¼ 1.7). The PC was used between several times a day and daily by this group (M ¼ 1.4) and the use was also estimated as rather easy (M ¼ 1.8). 7.2 Procedure As in Experiment 1, participants first answered the questions regarding their previous experiences with technical devices and completed the questionnaire assessing their LOC interacting with technology. According to the results in this questionnaire users were divided into two groups by median split. Then, they processed the same four tasks with the children twice consecutively, having a maximum of five minutes per task. 7.3 Results In a MANOVA performance, outcomes with regard to the number of tasks solved, the time needed to process the eight tasks and the inefficient keystrokes carried out in this process were calculated. Independent variables were the three key solutions and users’ LOC interacting with technology. A significant main effect was found for the key solution (F (6.76) ¼ 2.3; p 5 0.05), to be seen in figure 6. As visualised in figure 6, participants using phone C solved only 7.4 of the eight tasks, whereas phone A and phone B groups solved, on average, 7.9 tasks. Regarding the time needed, phone A showed the best performance with 430.3 s and phone C again the worst with 587.8 s. The performance in phone B group ranged in between with 459.7 s. With respect to the inefficient keystrokes, the largest performance difference between the groups was found with an average 6.1 inefficient keystrokes in the phone C group, 3.1 inefficient keystrokes in phone B and only 0.2 inefficient keystrokes in phone A. When looking at the single F-tests, statistical significance was reached for the number of tasks solved (F (2.39) ¼ 2.81; p 5 0.1) and for the number of inefficient keystrokes (F (2.39) ¼ 5.59; p 5 0.01). The main effect of LOC did not have a significant influence on users’ performance. Low LOC participants accomplished 7.5 tasks in 534.6 s and made 4.4 inefficient keystrokes. In contrast, the high LOC group solved 7.9 tasks in 452.5 s and made 1.9 inefficient keystrokes. The interaction between key complexity and LOC did also not reach statistical significance. Even though young adults compared to the children of Experiment 1 pressed fewer keys inefficiently, it is worth looking at which specific keys caused most trouble. Again, three examples of keys differing in the sense that they have semantically similar or dissimilar modes are looked at. 7.3.1 Key level 1. Key with two but semantically similar functions: The c-key of phone A one, which is used to return to higher levels in the menu hierarchy and to correct was never used inefficiently by the students. 2. Key with one but inconsistent function: The green receiver key of the two Siemens phones, which is used to effect calls, but does not have any function at other points in the menu was used 18 times in total when using phone C and seven times in total when using phone B. Figure 6. Effects of key complexity on the number of inefficiently used keys (left), the number of correctly solved tasks (centre) and the time on task (right) for the young adults group (N ¼ 45). On keys’ meanings and modes: children’s efficiency using a mobile phone 3. Keys with many and semantically dissimilar modes: The softkeys of the two Siemens key solutions exert many different functions depending on the point of the menu. The phone C keys, having many semantically dissimilar functions were hit 68 times in total and phone B softkeys with fewer incongruent functions were used inefficiently 22 times. 4. Keys not helpful for tasks solution: The students did not use the hash and asterisk keys. 7.3.2 Correlations between the dependent variables. In order to estimate what kind of relationship exists between the LOC of a user interacting with technology and other experiences with technical devices, correlations between LOC and users’ answers to questions concerning their technical expertise were correlated (Spearman-Rho for ordinal and Kendall-Tau-b values for nominal data). The highest correlation of r ¼ 7.57 (p 5 0.01) was found between LOC and the estimated ease using a PC, indicating that the higher participants’ LOC, the easier they find using a PC. The second highest correlation showed that the higher students’ LOC, the higher the students’ estimated interest in new technologies was (r ¼ 0.53; p 5 0.01). Furthermore, the higher participants’ feeling of having control over technical devices, the higher is their frequency of using a PC (r ¼ 0.52; p 5 .01) as well as a videocassette recorder (r ¼ 0.30; p 5 0.05). Between LOC and the frequency of using a mobile phone (r ¼ 0.12; n.s.) and the estimated ease of using one (r ¼ 70.13; n.s.), no significant correlations were found nor with users’ mobile phone expertise in terms of possession of a phone (r ¼ 0.00; n.s.) or the number of different functions used (r ¼ 0.01; n.s.). For further insight into the correlates of LOC interacting with technology, interrelations of age and gender as well as 427 performance interacting with the mobile phone were calculated. There was indeed a significant correlation of r ¼ 70.38 (p 5 0.01) between LOC and gender, indicating that male participants showed higher values in internal LOC and participants showed a somewhat higher LOC, the older they were (r ¼ 0.20; p 5 0.1). With regard to the LOC and performance interacting with the mobile phone, there was only a rather small correlation with the number of tasks solved (r ¼ 0.27; p 5 0.1), but insignificant correlations with time on task (r ¼ 70.20; n.s.) and inefficient keystrokes (r ¼ 70.07; n.s.). Gender correlated by r ¼ 0.25 (p 5 0.1) with time on task (women needed slightly more time), but not with any of the other performance measures. 8. Comparison of the predicted complexity and performance outcomes for children and young adults According to the three models proposed in section 4, differences in cognitive complexity of the three key solutions and relative decreases in users’ performance were predicted (figure 8, left side). Performance outcomes from the two experiments are now contrasted for the number of inefficiently used keys, as this was the most direct (different key complexities were examined) and accordingly, the most sensitive measure. In figure 7, the prediction (figure 7, left side) and the empirical values for children’s (figure 7, centre) and young adults’ (figure 7, right side) is shown (both user groups appear in different graphs, as the scaling of the axes was not compatible). The key solution with the lowest predicted complexity (phone A) is always set as the baseline and the higher complexity of the other solutions is calculated as increase compared to the baseline. From figure 7 illustrating the increases with respect to the number of inefficient keystrokes, it can be seen that Figure 7. Comparison of the predicted and the empirical outcomes: Number of inefficiently used keys in the three key complexities in children (left side) and young adults (right side). 428 M. Ziefle et al. performance decreased dramatically in phone B and phone C. This was predicted in all models (cf. table 2 and figure 1). However, model three, especially stressing the importance of dissimilar modes, proved to be the best predictive approach, even though the empirical results distinctly outreached the predictions. It was assumed that phone B would be about 167% more complex than phone A and phone C, even 244% more complex. Young adults, though, made 1450% more keystrokes with phone B and 2950% with phone C. Children had an increase in the number of inefficient keystrokes of 103% with phone B and even 335% with phone C. Admittedly, the absolute percentages are of course ‘virtual’, as the absolute weight of each of the factors assigned to in the models was hypothetical and on the other hand, they depend on the initial value of the phone A condition (since this is the basis the increases were related to). The model three, however, and this is the central outcome, had the best fit with the data, which means that the assumption of model three, that dissimilar modes add considerably more to complexity of a key solution than only the number of key options or modes, was shown to be true, and not only for children but also for young adults, who are bright and technology-prone. 9. Discussion The question of how to design control keys for mobile phones in order to be easily usable was addressed by two experiments with children and young adults as user groups. The 36 children and 45 young adults solved four common tasks with simulated mobile phones that had an identical menu, but differed in the complexity of keys to be used when handling the phones. As effects of keys’ complexity were isolated from the complexity of a menu, the difficulty using the mobile phone can be interpreted solely as a function of the different implementations of the control keys. The outcomes draw a clear and insightful picture and are now discussed with respect to their implications for the design of mobile phones. 9.1 Effects of keys’ complexity As was shown in the two experiments, the complexity of keys is indeed a critical factor affecting the efficiency using a mobile phone for a broad user group. Primarily, key solutions having modes with semantically incongruent functions were shown to be responsible for user’s inefficiency using the keys correctly. This is taken from the highest number of inefficient keystrokes found in the phone C key solution, which possessed the highest number of semantically inconsistent functions per key. Even if the absolute number of inefficiently used keys was different in the two user groups with children using the keys more often than younger adults did, child users made 4.4 times and young adults 30 times as many inefficient keystrokes with phone C keys than with phone A keys, the ones with the lowest complexity in both, number of key options and modes, with none of the modes semantically dissimilar. The mere number of key options is only of secondary importance, as phone B (comparable number of keys in phone but less keys with semantically inconsistent functions) performed distinctly better. Thus, it can be concluded here, that the keys’ complexity aggravates users’ effectiveness and efficiency, mainly due to the meanings of keys: if a single key has many meanings, with contradicting semantics, the cognitive demand for users is unnecessarily high, even in the second trial, where no learnability effects were found, as was shown in the phone C condition. However, the negative effects of complex key solutions were not limited to the number of inefficiently used keys, but also showed up in the time needed to process the tasks and the number of successfully solved tasks. Children were 10% more successful and 14% faster when using the phone A keys than the phone C keys, with phone B phone ranging in between. Similarly, the young adults solved 6.8% more tasks correctly with phone A keys and were 37% faster compared to the more complex phone C users. It is worth mentioning that a key’s mode itself, when semantically similar, did not deteriorate performance as strongly as for example in the ‘C’ key of phone A exerting two functions (corrections and returns in the menu), both meaning ‘undo’. Even if the children misused this mode key in the first trial, they learned to use the key in the second time where they did not use it incorrectly anymore. One of the phones (phone C) had redundantly implemented functions. According to the literature (e.g. Sanders and McCormick 1993), this should increase users’ performance as it provides a higher chance to match the user’s expectations of how a function is executed. As the children, though, did not detect the additional functionalities at all (they never used the arrow keys left and right, and only three of the young adults did so in more than two tasks) these keys do not have any identifiable effect on the usability of the phone. A final remark concerns the two age groups examined here. The results showed that the common prejudice of kids being superior in all technical complexities is not true, at least not in the most stringent form. If manufacturers (one of the phone manufacturers expressed that in a personal note to one of the authors) expect usability problems to vanish into thin air, assuming that future users groups – as they grow up with technology – will not have severe difficulties using technical products, these hopes will not easily be fulfilled. As demonstrated here, the still increasing (and not decreasing) importance of usability concerns in technical products should not be underestimated and the ‘biggest barriers of mobile computing’ (York and Pendharkar 2004) are still the ‘ergonomics and usability’ of On keys’ meanings and modes: children’s efficiency using a mobile phone 429 a technical device. Thus, if interfaces are not well defined and suboptimally designed, implying high cognitive demands on users, performance is very sure to be worse, independently of any characteristics of the user group. This finding confirms earlier results showing that suboptimal mobile interfaces caused a lower performance than welldesigned ones, independent of users’ age and expertise (Ziefle 2002b; Bay and Ziefle 2005; Ziefle and Bay 2005). This is especially crucial as the complexity of the keys present in some of the real devices was shown to be indeed not necessary (with respect to demands of software aspects), thus it is possible to implement a less complex one! 9.2 Effects of locus of control The felt competence in using different technological devices is another main factor predicting the ease using a mobile phone, especially for children that are commonly assumed to handle technical devices easily. The findings were indeed revealing as nearly any study has been concerned with the effects of this cognitive style in children when interacting with technology. The results confirm earlier results (Bay and Ziefle 2003b; Ziefle et al. 2004; Bremen and Ziefle submitted), showing that the reported competence using technological devices interacts with performance of adults. Briefly, two main points are important. Firstly, 9 – 14-year olds reporting to have low locus of control handling technology are sensitive enough to estimate their own capability to handle mobiles, which means they show considerably lower performance than the rest. Secondly, children feeling low confidence using technology had extreme trouble handling the complex phone C keys. Children with low LOC values used the keys three times as often inefficiently, solved on average one task less and were 26% slower when processing the tasks compared to children with high LOC values. Young adults with low LOC values were 5% less successful in task solving, needed 18% more time on task and made twice as many inefficient keystrokes, however, the LOC effects on performance were not found to be statistically significant. In order to find out why, a closer look is concerned with the absolute LOC values in both user groups. In figure 8, the distributions of children’s LOC (left) and young adults LOC (right) are visualised. As can be seen, the distributions of children’s and adults’ LOC values are very similar and therefore may not serve as an explanation of the lack of significance in the interaction of the degree of LOC and performance in the adult group. Focusing on correlation outcomes, the reported frequency of using technical devices (especially the PC) and the ease of using them was found to be interrelated with LOC values, however, for children and for young adults. On the basis of the present results the difference between the importance of the LOC concept for both user groups cannot be finally determined. However, one might speculate that the young adults examined here, being all university Figure 8. Distribution of the LOC values [max ¼ 100] in the children group (left) and the young adults group (right). students, might have compensated the LOC effects by their general high-cognitive abilities. Thus, in future studies the role of this cognitive style is to be studied in a less sapient user group, better representing the average user confronted with the complexity of technological devices. 9.3 Reviewing the models predicting keys’ cognitive complexity With the goal of defining the factors that contribute to cognitive complexity of a key solution for a menu-driven technical device and predicting users’ difficulties interacting with it, we have defined three models in this paper. The first model only included one factor, namely the number of different options where a user can stroke the keys of the device at each point of the task solution process. The underlying idea of this very simple model is the fact that a user has to make decisions concerning which of the keys s/he wants to press in each step to reach a certain goal. The more options are available, the more difficult the decision is. This model leads to the prediction that the phone B keys (eight options) will prove to be the most complex solution, phone A (four options) the easiest and the phone C (seven options) ranging in between these two. The second model introduced – additionally to the number of key options – two other factors: modes and redundancy. Modes are supposed to increase complexity because the different functions a key exerts need to be kept in mind, increasing the user’s cognitive load. Considering the total number of different functions each of the keys can exert and subtracting keys with redundant functions (which decrease cognitive load), it was predicted that the phone C keys would be about 150% and the phone B 100% more complex than the keys of phone A. The third model finally did not treat all kinds of modes as equally difficult for the user, but differentiated between semantically similar functions (basic modes) and 430 M. Ziefle et al. functions of a key that have completely different meanings (dissimilar modes). Dissimilar modes are supposed to be more cognitively demanding because they inhibit the formation of action schemas and are therefore weighed by the factor 2. This model predicted a 167% higher complexity of phone B compared to phone A and a 244% higher complexity of phone C compared to phone B. The performance data of our two experiments showed, in terms of inefficient keystrokes, increases in complexity with phone B of 103 % (children) and 1450% (students) compared to phone A. In the phone C condition increases were around 335% (children) and 2950% (students) in relation to baseline (phone A). Thus, the model that predicts the empirical data best, is model 3, considering key options, modes and meanings. This model can be regarded as a good approach for designers of key solutions on small menudriven devices who need to estimate how many difficulties users will have when interacting with the keys and it will help avoiding high cognitive complexity. However, the impact of keys’ meanings and modes on the complexity of a key solution seems to be underestimated, considering the fact that in both user groups the decrease in performance from phone B to phone C in terms of inefficient keystrokes was much higher than predicted. Especially semantically dissimilar modes may have contributed to complexity, because the phone C key solution has 37.5% more dissimilar modes than phone B and only 11% more modes in general, which may, in part, explain the big difference in performance between phone B and the phone C. Which other factors may have added to complexity and should be incorporated into a model? Concerning the modes, no differentiation was made between keys that exert one function at one point in the menu and no function at other points, and those keys that exert semantically dissimilar functions at different points of the menu. It can be supposed that there is a difference in the cognitive demands of those different situations. With regard to the fact that users should be enabled to establish action schemas for each of the keys, the allocation of semantically dissimilar functions to one key should definitely be avoided. Yet, whenever a key exerts its function only at very few points in the menu, having no function at all the rest of the time, it may be designed accordingly in terms of its appearance (size, unambiguous icon) and spatial location, as it was done with the green receiver key in phone B. This leads to a second important point, the spatial location of single keys within the other keys. The spatial allocation of keys has to be chosen according to their importance. It was shown in this study that two keys exerting exactly the same function, but one being allocated centrally and one laterally, lead to different frequency of misuse. As this green receiver key has to be used only rarely (only when effecting calls) whilst being centrally positioned, this leads in combination with the green colour to a misinterpretation as a confirmation key and therefore to many inappropriate operations. A third important aspect that may explain the great difficulties experienced by participants using the phone C solution are the rocker switches, which at times have two stroking options and at times exert only one function. This confuses users in addition to the many different semantic meanings these keys have, as was shown by a comparison of the keys with the softkeys of phone B. 9.4 Limitations of the study The present study represents a first attempt to propose and evaluate a quantitative model predicting the cognitive complexity of key solutions on small menu-driven devices, which consists of several contributing factors. Our model, comprising the number of key options, the modes and the meanings of the keys was able to predict performance using three different key solutions fairy well. Nevertheless, many more studies are necessary to evaluate the generalisability and further develop the model. In the previous section we have already mentioned three aspects that should be incorporated into the model: the different characteristics of semantically dissimilar modes, spatial locations of keys and double assignment of functions of one key, even though quantifying their contribution to complexity will not be easy. For reasons of ecological validity we chose three key solutions of existing (and widespread) mobile phone models. Unfortunately, the three factors included in the model could not be experimentally isolated from each other. In order to prove the generalisability of the model it will therefore have to be applied to a number of other key solutions, as well as to other menus and other devices. Furthermore, the fit of the model’s predictions with performance data of older adults will have to be verified. As this user group is especially sensitive to bad design, as was shown in a number of studies (e.g. Ziefle and Bay 2005), this should prove to be very insightful. 9.5 Recommendations for the design of mobile phones’ keys From the results of the two studies illustrated in this article, a number of recommendations for the design of control keys of mobile phones can be proposed: 1. Whenever modes cannot be avoided due to the limited space available on the device, semantically similar functions should be allocated to a key. This should enable the user to build up action schemas, which can be triggered autonomously, decreasing cognitive load. 2. Keys that exert their function only at very few points of the menu should be avoided or placed laterally on the device, unambiguously indicating its functions. If the very specific function of a key is not transparent, On keys’ meanings and modes: children’s efficiency using a mobile phone users may think it is broken, when it does not lead to any visible changes in the display. 3. The alternating allocation of one and two functions at a time to one and the same key should definitively be avoided. The user does not know where to press the key and makes many mistakes. 9.6 Research in progress Current research regarding the usability of navigation key solutions deals with transferring the effects found here to older adults as well as to devices with different types of menus. This will help to generalise the logic applied here to predict the complexity of keys to other devices, such as navigation systems, photocopying machines, digital cameras, wrist watches – basically all devices disposing of a hierarchical menu structure that has to be operated with a small number of keys. Acknowledgements We would like to acknowledge the valuable contribution of Alexander Schwade, who collected and analysed the children data. Sadly, he passed away before this paper was concluded. His inspiration will never be forgotten. Moreover, we thank Philipp Brauner for programming the software for displaying the mobile phones and logging user actions as well as Sarah Hatfield, Preethy Pappachan and Judith Strenk for their research support. A final thanks is devoted to two anonymous reviewers for their constructive comments on an earlier version of this manuscript. References BAUMANN, K., 2001, Controls. In User Interface Design for Electronic Appliances, K. Baumann and B. Thomas (Eds), pp. 131 – 161 (London: Taylor & Francis). BAY, S. and ZIEFLE, M., 2003a, Performance on mobile phones: Does it depend on proper cognitive mapping? In Human Centred Computing. Cognitive, Social and Ergonomic Aspects, D. Harris, V. Duffy, M. Smith and C. Stephanidis (Eds), pp. 170 – 174 (Mahwah, NJ: Lawrence Erlbaum). BAY, S. and ZIEFLE, M., 2003b, Design for all: User characteristics to be considered for the design of phones with hierarchical menu structures. In Human Factors in Organizational Design and Management – IV, H. Luczak and K.J. Zink (Eds), pp. 503 – 508 (Santa Monica: IEA). BAY, S. and ZIEFLE, M., 2004, Effects of menu foresight on information access in small screen devices. In Proceedings of the 48th Annual meeting of the Human Factors and Ergonomic Society (Santa Monica: Human Factors and Ergonomic Society), pp. 1841 – 1845. BAY, S. and ZIEFLE, M., 2005, Children using cellular phones. The Effects of shortcomings in User Interface Design. Human Factors, 47(1), pp. 158 – 168. BEIER, G., 1999, Kontrollüberzeugung im Umgang mit Technik [Locus of control regarding the use of technology]. Report Psychology, 9, pp. 684 – 693. BREMEN, K. and ZIEFLE, M., submitted, Effects of cognitive and personal factors on PDA navigation performance. International Journal of Human-Computer Studies. 431 BURMESTER, M., 1997, Guidelines and Rules for Design of User Interfaces for Electronic Home Devices. The Esprit Project 6994 (Stuttgart: Fraunhofer IRB). COOPER, A., 1999, The Inmates are Running the Asylum. Why high tech products drive us crazy and how to restore from sanity (Indianapolis: Sams). HELLE, S., JÄRNSTRÖM, J. and KOSKINEN, T., 2003, Takeout menu. The elements of a Nokia users interface. In Mobile Usability. How Nokia changed the face of the mobile phone, C. Lindholm, T. Keinonen and H. Kiljander (Eds), pp. 47 – 71 (New York: McGraw-Hill). JORDAN, P.W., 1998, An Introduction to Usability (London: Taylor & Francis). KIERAS, D. and POLSON, P.G., 1985, An approach to the formal analysis of user complexity. International Journal of Man-Machine Studies, 22, pp. 365 – 394. KIM, H. and HIRTLE, S., 1995, Spatial metaphors and disorientation in hypertext browsing. Behaviour and Information Technology, 14, pp. 239 – 250. LEE, S. and HONG, S.H., 2004, Chording as a text entry method in mobile phones. In Mobile Human-Computer-Interaction – MobileHCI 2004. S. Brewster and M. Dunlop (Eds), pp. 456 – 460 (Berlin: Springer). LIN, D-Y., 2001, Age differences in the performance of hypertext perusal. Proceedings of the Human Factors and Ergonomic Society 45th Annual Meeting (Santa Monica, CA: Human Factors and Ergonomics Society), pp. 211 – 215. MARTEL, A. and MAVROMMATI, I., 2001, Design principles. In User Interface Design for Electronic Appliances, K. Baumann and B. Thomas (Eds), pp. 77 – 107 (London: Taylor & Francis). NORMAN, D.A., 1981, Categorization of action slips. Psychological Review, 88, pp. 1 – 15. PAK, R., 2001, A further examination of the influence of spatial abilities on computer task performance in younger and older adults. Proceedings of the Human Factors and Ergonomic Society 45th Annual Meeting (Santa Monica, CA: Human Factors and Ergonomics Society), pp. 1551 – 1555. RASKIN, J., 2000, The Humane Interface. New directions for designing interactive systems (Reading, MA: Addison-Wesley). SANDERS, M.S. and MCCORMICK, E.J., 1993, Human Factors in Engineering and Design, 7th edition (New York: McGraw-Hill). SHNEIDERMAN, B., 1998, Designing the User Interface (Reading, MA: Addison-Wesley). VICENTE, K.J., HAYES, B.C. and WILLIGES, R.C., 1987, Assaying and isolating individual differences in searching a hierarchical files system. Human Factors, 29, pp. 349 – 359. WEISS, S., 2002, Handheld Usability (Chichester, West Sussex: John Wiley & Sons). YORK, J. and PENDHARKAR, P.C., 2004, Human computer interaction issues for mobile computing in a variable work context. International Journal of Human Computer Studies, 60, pp. 771 – 797. ZIEFLE, M., 2002a, Usability of menu structures and control keys in mobile phones: A comparison of the ease of use in three different brands. Proceedings of the 6th International Scientific Conference on Work With Display Units, (Berlin: Ergonomic) pp. 359 – 361. ZIEFLE, M., 2002b, The influence of user expertise and phone complexity on performance, ease of use and learnability of different mobile phones. Behaviour and Information Technology, 21, pp. 303 – 311. ZIEFLE, M. and BAY, S., 2005, How older adults meet complexity: Aging effects on the usability of mobile phones. Behaviour and Information Technology, 24(5), 375 – 389. ZIEFLE, M., BODENDIECK, A. and KUENZER, A., 2004, The impact of user characteristics on the utility of adaptive 5help systems. In Work with Computing Systems, H.M. Khalid, M.G. Helander and A.W. Yeo (Eds), pp. 71 – 76 (Kuala Lumpur: Damai Sciences, 2004).