Search tactics used in solving everyday how-to technical tasks: Repertoire, selection and tenacity

With greater access to computational resources, people use search to address many everyday challenges in their lives, including solving technology problems. Although there are now many useful ‘how-to’ resources online (especially videos on YouTube), it can still be diicult to identify, understand, and resolve certain kinds of technical problem. While research tasks have been studied for many years and we know the tactics people use, we know far less about searchers’ tactics for how-to technical tasks that involve actually being able to apply found information to resolve a problem. Crucial to our study was developing and studying a highly realistic, how-to technical task, for which there was no single guidance resource: making a phone safe for a child. After providing 39 participants with an actual phone to ix, and a search engine to perform the task, we analysed their search tactics using retrospective cued think aloud interviews. Our primary contribution is a set of 77 tactics used, in three categories, along with detail of how common they were. We conclude that people had a lot of tactics in their repertoire. Although it was not hard for participants to ind relevant information, what was hard was for participants to ind information they could use; indeed only 23% of participants successfully completed the entire task. Domain knowledge afected the choice of tactics used (although not necessarily to- wards better task success). We discuss these inluences and make design recommendations for how future search systems can support those in resolving how-to technical tasks. and Joo (2010) have extended Bates’ groups of tactics to include evaluation of the search results. Although not calling them tactics, Barry and Schamber's (1998) evaluation criteria collected from both university students and staf, as well as occupational users of weather-related information is more extensive. Building on this study, Savolainen and Kari (2006) investigated the use of these criteria in web search by asking their participants to search on a topic related to their own self-development such as a hobby. It seems likely that there is a universal set of tactics (Savolainen, 2017) but which tactics are employed is dependent on a number of factors including task, perceived knowledge, search system and process (Xie & Joo, 2012). It can be diicult to make comparisons across studies as the approaches taken to identifying tactics difer and tactics are identiied at varying levels of granularity. Tactics can either be identiied from a person's explanation of the action taken (e.g. Bates, 1979b) or by analysing chronological sequences of moves (e.g. Thatcher, 2006; Wildemuth, 2004). Studies may identify a detailed list of a particular group of tactics / moves (e.g. Fidel, 1985) or just a range of tactics at a high level (e.g. Xie & Joo, 2010). In our study, we are interested in people's explanations of all the actions taken while searching to resolve an information problem. Understanding how people make sense of their own searching activities can help in uncovering additional features that should be taken into account in design interventions, whether that involves designing better interfaces and functionality for search engines or designing better learning experiences to help people search more eiciently and efectively.


Introduction
Many everyday problems can be solved by a simple one-shot search, where a person types a query into a search engine, selects one of the top-ranked results, and obtains the information they need to address their problem. There are many useful how-to resources, for helping people complete everyday tasks like cooking and DIY, especially on YouTube (Torrey, Churchill, & McDonald, 2009). Nevertheless, there are problems that require more efort to try and obtain useful and usable information. When we cannot ind a simple answer to a technical problem, it can be challenging for even technically minded individuals to solve. We were interested in how people dealt with these more challenging tasks, and the tactics used. This kind of task can involve iterative searching, and also iterations between searching and acting with the results, and so we expected to see additional tactics that lead to task success. Notably, success is measured not by the relevance of the information retrieved, but by whether the information retrieved helps a person to actually solve the problem in their subsequent interaction with the technology.
Our objective was to identify search tactics for resolving everyday how-to technical tasks, in order to identify ways in which search systems can help struggling searchers. Through this work, we contribute: a) A novel characterisation of information-seeking when resolving everyday how-to technical tasks, as opposed to open ended research tasks. b) An expanded set of tactics that build on prior work. c) Evidence that everyday users have lots of tactics in their repertoire; they don't necessarily need to be taught 'more tactics'. d) Evidence that the task type invoked diferent types of tactics to research tasks in previous literature. e) Recommendations for how future research, and search systems, can support those resolving everyday how-to technical tasks.
We conclude that to improve people's search skills, we should focus on helping them understand types of tasks, and appropriate tactics for them, rather than on teaching them new ones. Bates (1990) deines a move as the basic unit of analysis -"an identiiable thought or action that is part of information searching". A tactic is "a move or moves made with the purpose of improving or speeding the search in some way" i.e. a tactic is a grouping of moves that may incorporate both thought-moves and action-moves. These deinitions continue to be a basis for many subsequent studies. In our study, we are interested in search tasks where the searcher must use the information found to resolve how-to technical problems. A search can only be considered successful if the information found does actually resolve the information problem. We, therefore, extend Bates ' (1990) deinition of a tactic to "a move or moves made with the purpose of improving or speeding the search in some way to resolve an information problem".

Search tactics
The focus of our study and this review is tactics; the "irst level at which strategic considerations are primary" (Bates, 1990), as it is here that the cognitive approach taken to search is revealed (Hsieh-Yee, 1993). It should be noted that for many studies the focus is on moves (e.g. Fidel, 1985;Wildemuth, 2004;Zhang, Jansen, & Spink, 2006) . We review these studies too if tactics are also discussed (Table 1). Some studies also try to identify tactics from system logs. He, Qvarfordt, Halvey, and Golovchinsky (2016), for example, attempted to automatically identify tactics in system logs, but had to focus on a small number of (e.g. 6) tactics that had explicitly observable user actions (or 'moves') in them. The broader sense of tactics we are concerned with involve moves or actions that could be attributed to multiple diferent tactics, and ultimately require a cognitive form of investigation to identify.
In a seminal study, Bates (1979a,b) identiied tactics based on her own experience and that of professional information specialists searching subject databases. Bates (1979a,b) identiied ive groups of tactics (1) Monitoring tactics -keeping the search on track, (2) File structure tactics -ways of locating items in search systems, (3) Search formulation tactics -design of the search query, (4) Term tactics -selection and organisation of query terms, and (5) Idea tactics -ways to resolve search problems. Also examining how professional information specialists search, both Fidel (1985) and Shute and Smith (1993) considered the moves and tactics made in online searching and found that these largely correspond with the tactics identiied by Bates (1979a,b) but that while all Bates' tactics are theoretically possible, professional searchers only use some (Fidel, 1985).
Since these studies of professional searchers, the rise of the web has meant that much in the information environment has changed: full text is indexed, content is uncontrolled, results are ranked, and search is an everyday activity for many people, not just professionals (Mlilo & Thatcher, 2014). Subsequent studies have considered the tactics of web users (Savolainen & Kari, 2006;Smith, 2012;Thatcher, 2008;Xie & Joo, 2010). Many of the tactics identiied by Bates (1979a,b) have also been observed in these studies (Hsieh-Yee, 1998;Smith, 2012) but some new tactics and tactic groups have also been identiied. Most notably both Smith (2012), and Xie and Joo (2010) have extended Bates' groups of tactics to include evaluation of the search results. Although not calling them tactics, Barry and Schamber's (1998) evaluation criteria collected from both university students and staf, as well as occupational users of weather-related information is more extensive. Building on this study, Savolainen and Kari (2006) investigated the use of these criteria in web search by asking their participants to search on a topic related to their own self-development such as a hobby.
It seems likely that there is a universal set of tactics (Savolainen, 2017) but which tactics are employed is dependent on a number of factors including task, perceived knowledge, search system and process (Xie & Joo, 2012). It can be diicult to make comparisons across studies as the approaches taken to identifying tactics difer and tactics are identiied at varying levels of granularity. Tactics can either be identiied from a person's explanation of the action taken (e.g. Bates, 1979b) or by analysing chronological sequences of moves (e.g. Thatcher, 2006;Wildemuth, 2004). Studies may identify a detailed list of a particular group of tactics / moves (e.g. Fidel, 1985) or just a range of tactics at a high level (e.g. Xie & Joo, 2010). In our study, we are interested in people's explanations of all the actions taken while searching to resolve an information problem. Understanding how people make sense of their own searching activities can help in uncovering additional features that should be taken into account in design interventions, whether that involves designing better interfaces and functionality for search engines or designing better learning experiences to help people search more eiciently and efectively.

How-to technical tasks
A task is a goal-based activity (Hackos & Redish, 1998), that may consist of sequences of sub-tasks that need to be performed to achieve an outcome (Toms, 2011). There are both work tasks and search tasks. A work task is "an activity people perform to fulil their responsibility for their work" (Li, 2009). The concept of a work task is not limited to the work arena and the term can also be used to describe activities within a leisure context (Vakkari, 2003). Many of the studies of tactics have either been for research type work tasks (Bates, 1979a,b;Fidel, 1985;Shute & Smith, 1993;Vakkari, Pennanen, & Serola, 2003;Wildemuth, 2004) and/or for work tasks within particular domains (Bhavnani, 2001;Xie & Joo, 2012). For example, Xie and Joo (2012) compare popular, occupational and scholarly work tasks. It is likely that diferent types of work task co-exist within each group. In this study we are interested in a particular type of work task: everyday how-to work tasks to ix a technical problem.
Everyday tasks are those that "people employ to orient themselves in daily life or to solve problems not directly connected with the performance of occupational tasks" (Savolainen, 1995, p. 267), although in some circumstances and for some people these tasks Table 1 Moves and tactics identiied in prior work.
The author & information professionals Barry and Schamber (1998) (1) 23 evaluation criteria categorised under 7 groups: Information contents of documents, Sources of documents, Document as a physical entity, Other information or sources within the environment, Users situation, User's beliefs and preferences, User's previous experience or background.
(1) University students and staf (2) 32 evaluation criteria categorised under 10 groups: Accuracy, Currency, Speciicity, Geographic proximity, Reliability, Accessibility, Veriiability, Clarity, Dynamism, Presentation quality (2) Professional users of weather information Bhavnani (2001) 5 operators: Find-websites, Scan-websites, Compare, Verify, End-task (1) Healthcare experts (2) Shopping experts Fidel (1985) 18 operational moves and 12 conceptual moves Professional searchers Hsieh-Yee ( Web users with varying knowledge and experience, conducting 4 types of search task: 2 fact-inding (1 dispersed structure, 1 category structure) and 2 exploratory (1 dispersed structure, 1 category structure). Savolainen and Kari (2006) 18 user-deined relevance criteria Web searchers self-chosen topic related to self-development Shute and Smith (1993) 13 knowledge based tactics categorised into 3 groups: Broaden topic scope, Narrow topic scope and Change topic scope Expert intermediaries conducting literature searches Smith (2012) 34 Internet search tactics categorised into four groups: Monitor, File structure, Formulation and Term Internet search tactics identiied by author from research literature Thatcher (2006Thatcher ( , 2008 28 moves categorised into ive broad areas: Task initiation behaviours, Search terms, Sustaining search behaviours, Terminating behaviours, and Unusual search behaviours. Web users with varying experience searching for 2 researcher and 2 participant deined tasks that are either fact-inding or general purpose browsing 78 tactics identiied from chronological sequences of search moves 12 cognitive search strategies identiied from search behaviour patterns Vakkari et al. (2003) 6 search formulation tactics and 4 other tactics Students with varying IR knowledge conducting searches for 1 work task: a research paper Wildemuth (2004); 13 moves categorised under 5 groups: Beginning moves, Moves to reduce the size of the set, Moves to increase the size of the set, Moves to increase both precision and recall, Other moves University students conducting a database search for clinical problems in microbiology at three time points.
Tactics identiied from sequences of moves Joo (2010, 2012) 13 tactics a : Identifying search leads to get started, Creating search statement, Modifying search statement, Evaluating an individual item, Evaluating search results, Keeping a record, Accessing forward, Accessing backwards, Learning, Exploring, Organising, Monitoring, Using / Obtaining Web search general public 1 work related task and 1 personal-related task Xie, Joo, and Bennett-Kapusniak (2017) (2) self-generated task (academic and personal) in four systems (online database, search engine, OPAC, digital library) a Tactics as reported in Xie and Joo (2010). may be job related (Savolainen, 1995). How-to tasks are where a person needs to ind out how to do or make something. For example, a person may want to know how to bake a cake or make a best man's speech. When a person lacks information for a work task, this might result in one or more search tasks (Vakkari, 2003). Although search tasks for how-to work tasks may involve many of the same search tactics as a research task (a task that involves inding out about a given topic, and assembling resources deemed relevant), it is likely that there will be diferences because of the task characteristics. While for a research task there might be many results that are Topically and Cognitively relevant (Saracevic, 2007), information use in how-to technical tasks means that results have to be Situationally (usable) and Afectively (successful) relevant to the searcher. We suggest that for everyday how-to tasks in the technical domain what counts as domain knowledge and success is diferent. Furthermore, what information is used for and who searches for this information is diferent.
For research tasks a person may or may not have domain knowledge ("knowledge of the subject area" -Wildemuth, 2004) depending on their familiarity with a topic. However, for everyday how-to technical tasks a person may be familiar with aspects of the task, regardless of whether they have or do not have technical knowledge. For the task in our study ("make a phone safe for a child"), familiarity with phone safety, the brand of phone used in the study, as well as technical knowledge are all types of domain knowledge. An approach used in many studies to understand how domain knowledge inluences tactic use is to compare the searching tactics of those familiar with a domain with those who are unfamiliar (Hsieh-Yee, 1993;Shute & Smith, 1993;Xie & Joo, 2012). Given the inevitably wide variation in elements of subject area knowledge for everyday tasks that are also within a specialist domain we can't easily categorise our participants by subject area knowledge binaries and so we cannot take this comparative approach in our study.
In studies of information-seeking, search success is only rarely linked to whether the searcher can achieve their work task, by resolving the actual problem that led to the use of search systems (Kelly, 2007;Wildemuth, 2004). In studies of research tasks, when success is determined, it is often linked to inding relevant documents and the number of solutions found during the search session (e.g. Vakkari et al., 2003), or task completion speed (Aula & Nordhausen, 2006). While these may be good measures for research-style tasks (Wirth, Sommer, Von Pape, & Karnowski, 2016), for a how-to task information may be relevant but unless the information can resolve the problem it may not be useful. We expect that this will lead to diferent tactic use.
Many studies of tactics have found that search expertise is a factor in tactic use (Hsieh-Yee, 1993;Navarro-Prieto, Scaife, & Rogers, 1999;Thatcher, 2008;Xie & Joo, 2012). As we are investigating everyday tasks, those undertaking such searches will likely have variable search skill. We could therefore expect a wide variation in tactic deployment for this type of task, depending on the searchers' expertise.
How information is used may also inluence the type of tactics employed. For how-to technical tasks, we suggest that information will be used (1) to get instruction -"to ind out how to make and do things" and (2) to extend knowledge "to ind out about a particular aspect of a topic" (Rutter, Clough, & Toms, 2019). Search task characteristics such as goal speciicity have been shown to inluence tactic selection in two studies (Navarro-Prieto et al., 1999;Xie & Joo, 2012) although not in two others (Hsieh-Yee, 1993;Thatcher, 2008).
To summarise, most studies of tactics are for research type work tasks or work tasks within a particular domain. In our study, we investigate search tactic use during how-to type work tasks. We anticipate that we will see diferences from other studies because of the task characteristics. For everyday how-to tasks (1) searchers may be familiar with a task regardless of their subject knowledge, (2) information found during a search may be topically relevant but unless it can be used may not actually resolve the task problem (3) those conducting such searches will likely have variable search skill and therefore we could expect a wide variation in tactic deployment (4) there will be diferent information uses (such as to get instruction and to extend knowledge) that will inluence tactic use.
With the growth of technologies in our work and home lives, and the growth of how-to technical help available online growing number of people are searchig online for soltions to problems that they are having with their technology use. Whereas in the past, the smaller numbers of people using technology would have called upon designated tech support experts in a work context, across many settings, both work and domestic, people are searching online as part of their help seeking activities. Sometimes this search is fast and unproblematic -the person types in a query and easily identiies a result that is relevant, trustworthy and actionable by them to solve their problem. But what about harder problems, when composing the query is less obvious, where iterated searches are needed, or where the results are harder to assess for relevance, or are simple too confusing for the searcher to make use of? Our aim is to build on established prior work in online searching by people with varying levels of expertise, but try and understand if there are diferences in the case of non-trivial tasks involving searching for relevant and actionable how to technical solutions.

Research design
To identify variation in search tactics, we planned our study to include realistic tasks designed to elicit diferent search tactics, and recruited participants with varying degrees of search and domain knowledge. Note that this is an exploratory approach, aiming to scope out the range of variation in the things that people do. Consequently, it is not about hypothesis testing or even measuring the relative contribution of diferent variables. Rather it is about getting an understanding of the range of tactics that should be taken account of in informing future analysis and systems design.

Design of the tasks
With an emphasis on high ecological validity, and real task consequence (Borlund, 1997), we aimed for a work task that was realistic and seemed plausible that participants would be asked to address it. Our chosen task setting, shown in Fig. 1, was to ind out how to make an Android phone safe for a child. This involved three subtasks. To keep the participants focused on actual solutions rather than relevant results, and in line with Borlund's (1997) recommendations for simulated work tasks, we asked our participants to implement changes on a provided phone. These tasks could be credibly assigned by "a friend" -indeed several of our participants stated unprompted that their friends typically ask them to do this type of task or that they typically ask their friends for similar help.
The three tasks within the larger task environment were intended to require participants to generate search tasks and search subtasks to complete. This would reveal diferent tactics, and were designed with diferent types of information use in mind that are typical for everyday technical problems. For tasks 1 and 2 participants needed to ind instructions and for task 3 they had to ind further information about phone safety. This meant that the speciicity of the search goal varied: for tasks 1 and 2 the goal is speciic as there is particular information that must be found whereas for task 3 the goal is open as potentially any information relating to the topic could be used. We also made sure that these tasks were (a) hard for technologically competent people and (b) not solvable with a single 'how-to' video on YouTube. No advanced technical skill was needed to implement solutions on the provided phone. The solution to task 1 is to set up parental controls in Play Store. This information was readily available online but to implement the solution participants needed to locate Play Store on the phone, and determine the appropriate age-rating. Task 2 is easy to resolve on most phones by changing the settings menu. However, it is not possible to set up separate user proiles on the particular version of the phone we provided. At the time of this study, solutions for other phone models appeared in the search results pages -for this reason we anticipated that the solution for the study phone may be hard to ind. As a research-style task, there are many solutions to task 3, that we anticipated would be relatively easy to ind. We expected that task 1 and task 3 would be easy to accomplish but task 2 more diicult. In our pilot test we found that the tasks were achievable but not too easy.
We also considered other aspects of the tasks. We carefully worded the task statement so that keyword terms such as "parental control" and crucial information such as the phone model were missing. We also deliberately sequenced the tasks illogically, as ideally task 2 should be completed irst as otherwise parental controls (task 1) will need set again for the new proile. We did this in part because we did not want to discourage participants if they got stuck on task 2, but also because we wanted to see if any participants would evaluate the whole task and then change the task order.

Domain knowledge and search skill
To identify the variation of search tactics used to resolve a how-to technical tasks, we wished to study participants with difering search skill and domain knowledge. To identify search skill, we asked our participants to rate themselves against two proiciencies relating to online search taken from the EU Digital Competence framework (Ferrari, Neža Brečko, & Punie, 2014). Similarly, we use the same framework to identify technical knowledge by asking our participants to rate themselves against two proiciencies relating to technical competence (see Appendix B). To identify task familiarity, we designed our own questionnaire based on what knowledge we thought participants could use when resolving these tasks (see Appendix B).

Participants
To recruit participants with varying domain knowledge and search skill, we used self-reported search skill and technical knowledge (see procedures below) in our recruitment strategy. Initially, we recruited participants to: "take part in a study about solving technical problems using a search engine". After determining that our initial sample of participants had mostly self-reported high search skill and high technical knowledge, we later recruited people with posters asking: 'do you ask other people to solve your tech problems?' as some of our already recruited participants with low search skill and low technical knowledge told us they often S. Rutter, et al. I n f o r m a t io n P r o c e s s in g a n d M a n a g e m e n t 5 6 ( 2 0 1 9 ) 9 1 9 -9 3 8 asked their friends for help with technical problems. We recruited 39 participants: 17 self-reported as having high search skill and high technical knowledge, 13 had high search skill and low technical knowledge, and 9 had low search skill and low technical knowledge ( Fig. 2). Because of the chosen domain, however, we did not have any participants that self-reported as low search skill and high technical knowledge, implying that having good "tech knowledge" came along with higher search skill in our sample. The participant group consisted mainly of young, well-educated people. 24 were under 25 years of age, eight were between 26-35, three were 36-45, and four were 46+. 14 were male and 25 were female. 13 participants were undergraduates, 8 were MSc students, 13 were doctoral students, four were employed, and one was unemployed. Most participants were from a range of diferent academic and non-academic departments in the university, and many participants were from diferent countries. Although less than half had English as their irst language, only two participants thought that language inluenced their search process.

Data collection procedures
After gathering informed consent, participants were presented with a pre-task questionnaire: a self-report of search skill and technical knowledge based upon the EU Digital Competence framework (Ferrari et al., 2014). Participants were then given 20 min to make progress on the task. We did not allocate speciic time periods to each task, and so participants could work towards the larger of making the phone safe by attempting the subtasks in any order, or indeed in combination. After the time was up, participants completed the task-domain familiarity questionnaire (given post-task so that participants were not primed), before reviewing their task performance, via screen recordings, during a post-task interview. This post-task interview allowed us to capture a relective cuedretrospective Think Aloud (van Gog, Paas, van Merriënboer, & Witte, 2005) of their search processes and gain insight into their cognitive activities. The study was approved by the School of Computer Science's ethics board at the University of Nottingham, and participants received a £10 Amazon Voucher as remuneration for their time. The browser and phone were reset between participants to remove revisitation indicators for subsequent participants. The web search results (live un-inluenced results provided by the search engine of the participants' choice) were likely to be inluenced by our academic network address and geographic location, however this location was consistent for all participants.

Tactic identification and coding
We used the post-task interview transcripts, supported by screen recordings of the phone and laptop, and video footage of the participant to identify tactics and knowledge used. We analysed all searching behaviour on both devices including the searches participants did to ind solutions as well as any search conducted when trying to implement solutions. For each participant their interview was transcribed verbatim. Tactics that participants described were then matched by the irst author with (a) the actions taken in the search user interface, and (b) what type of knowledge they were using, if any (Fig. 3).
During the irst round of coding, tactics were coded inductively (open coding) based on the participant's explanation of why they searched as they did including their query formulation, results evaluation and how they managed the tasks. In the second comparative deductive phase, the unique tactics in the codebook were primarily cross-referenced against tactics from two well-known sets of tactics produced by Bates (1979b) and Barry and Schamber (1998) (selective coding). Together, these covered a good range of search and evaluation tactics, if a tactic could not be cross-referenced, tactics were linked to other sources where possible. For rigour, this comparative approach was performed in both directions -considering our tactics against those listed in these prior works, and considering each of the tactics in prior works against our own data. As inter-rater reliability can only be carried out using a "simpliied coding scheme" (Patton, 2015, p. 667) and the irst author had identiied 77 tactics, we did not carry out a formal inter-rater reliability assessment. However, at group meetings all authors discussed and reined the coding scheme using example data provided by the irst author. This process produced (1) a series of interview transcripts, (2) an accompanying annotated table of tactics used by each participant, and (3) a single codebook of unique tactics, with cross-references to participants. We then considered tactic frequency. We suspected that participant fatigue meant that "new" tactics were described but the repeated use of the same tactic was not. For this reason, we report tactic frequency across our data set (how many participants used a tactic) rather than how many times each participant used each tactic. We report these igures in the appendices in each of the tactics tables. It should be noted that while this method, and analysis, provided very rich and insightful data about tactic use, the focus of the discursive interviews is to identify diferent types of tactics, not on providing comprehensive counts of every time a tactic is used.

Quantifying domain knowledge and search skill
We expected to see quantiiable diferences in tactics between those who self-reported as having technical knowledge, task familiarity and search skill. However, despite extensive analysis we could ind no meaningful diferences between the diferent knowledge groups. We also analysed the search logs for diferences (number of queries, length of queries, mouse hovers, number of results viewed, use of tabs, and so on), and only found one statistically signiicant result: those with low tech knowledge inspected web pages listed later in the search results (H: 4.83 (SD:2.23), L: 6.86 (SD:2.76), t(37) = 2.49, p = 0.02). We conclude that either selfreport data is unreliable (Hargittai, 2005) and/or that knowledge is too complex to quantify for this type of task. Therefore, in this paper, we primarily report our qualitative data and our participants' explanations for the tactics they selected. We plan to report our quantitative data in a related paper.

Task success
We asked participants to implement the solutions on the phone so that they could test whether the solutions they found worked. We only considered a search successful if the information found could be used to resolve the information problem. No technical skill was required to implement any of the solutions. The tasks were therefore a test of search skill.
As we allowed our participants to complete the tasks in any order, we irst identiied which of the tasks were started. We then checked (using the phone screen recording) whether participants had successfully resolved and implemented each of the tasks, and whether the solutions were correct or incorrect. We also report participants' explanations of why they stopped working on a task before it was completed.

Task success
Task success is reported in Table 2. All 39 participants started task 1. For 3 (8%) participants this was the only task they started, 8 (21%) participants started two tasks, and 28 (72%) attempted all three tasks. Overall 22 (56%) participants felt they did not have enough time to complete all three tasks. Only 9 (23%) of our participants completed all three tasks successfully. A further 21 (54%) found some solutions for task 1 and 3, but were unable to correctly resolve task 2 (although some thought they had). 8 (20%) participants were unable to ind any correct solutions. 1 participant found a correct solution for task 1 but could not implement the solution because they were unable to ind basic information such as the location of the phone's settings menu. As this participant could not implement the solution, we count this as unsuccessful. For many of our participants these were challenging tasks, particularly because solutions to other phones and models were presented in the search results that did not work for the phone provided in our study. S. Rutter, et al. I n f o r m a t io n P r o c e s s in g a n d M a n a g e m e n t 5 6 ( 2 0 1 9 ) 9 1 9 -9 3 8 Part of the reason why the success rate is so low is that 11 (28%) participants considered tasks complete even though they were aware that there was more they should or could do. These participants felt that they could stop doing the task when they had provided a partial solution. 2 (5%) participants reported that they gave up on tasks because they didn't understand them. 17 (44%) participants described not completing tasks because under time pressure when they couldn't ind information, they swapped to other tasks.

Search tactics for resolving how-to technical tasks
All bar one participant started the assigned tasks by searching for information. The only participant that started by directly trying to conigure the phone was not able to successfully achieve the desired result this way, and so all participants did search for solutions. Our analysis found three main types of search tactics: 1) 24 tactics to CONTROL the search -to direct what information is received and to manage information across multiple devices, 2) 29 tactics to SELECT and USE information to select which search results to visit, and what information is used in problem resolution, 3) 24 tactics to MANAGE the task and process -to answer the tasks and manage the search process, at a level that coordinated both the use of CONTROL tactics, and the use of SELECT and USE tactics.
These tactics, which are a primary contribution of the article, are extensively and comprehensively reported in Appendix A. In each table, we show how many of the participants reported using each tactic. In Table 3, we summarise tactic use. Individual participants used between 21 and 36 diferent tactics (mean = 28.6), out of a total of 77 tactics identiied. This suggests a large repertoire of tactics that people can draw upon. More tactics were identiied to SELECT and USE information (29) and MANAGE the task and process (24) than CONTROL the search (24). However, participants applied more control tactics (mean = 11.4) than SELECT and USE (mean = 9) and MANAGE (mean = 6.7) tactics.

Frequent CONTROL tactics
We next describe the CONTROL tactics that more than half of the participants used. For a full list of CONTROL tactics see Table 6, Appendix A. 35 (90%) participants described how they found terms to use in query formulation. This was an assigned task so not surprisingly 24 (62%) of these participants extracted terms from the task statement (C1). However, we had carefully worded the statement to avoid key terms: 28 (72%) participants realised that they needed additional information from another source (C5) and so looked for the model and operating system of the phone (either on the phone or on the laptop by searching for information about the phone) to use in query formulation and results examination. Participants also became aware of the importance of the model and operating system of the phone while searching. This inluenced what terms they used in query formulation. After inding information about the phone, 32 (82%) participants used terms relating to the phone model and operating system in query formulation (C9) in an attempt to restrict (unsuccessfully) the results to the type of phone. 20 (51%) participants described varying query terms (C12) because (a) they  S. Rutter, et al. I n f o r m a t io n P r o c e s s in g a n d M a n a g e m e n t 5 6 ( 2 0 1 9 ) 9 1 9 -9 3 8 could not ind relevant information in the search results and (b) as a technique to ind additional sources of information that could be used to conirm previous sources. However, much to our participants' frustration this tactic did not change the results much. Another common query formulation tactic used by 25 (64%) participants was to use exact terms (C10). This was because they sometimes needed to re-ind information on the other device, and so what was being looked for was already known. Another set of tactics was to locate terms on the phone / computer interface (C15 & C16). 33 (85%) participants needed to locate known terms on either the phone or laptop interface. Mostly this tactic was used to locate menus on the phone that had been described in the solutions found on the computer but it was also used to scan lengthy web pages. 27 (69%) participants also searched for what they thought could be alternative menu names (C16). They needed to do this because for task 2, the solution that many participants found described a menu that did not exist on the particular phone provided in this study.
All 39 participants used both devices simultaneously (C19) whereby they implemented solutions on the phone while viewing instructions on the laptop. This tactic is likely linked to this being an instruction-based task. When participants were unable to use this tactic, for example, when they found instruction on the phone and then had to move to a diferent screen to implement the solution they were frustrated. 30 (77%) participants also kept information available for future use (C17) by either opening new tabs or by taking notes.
Because the solutions were not easy to ind and particularly for task 2 where incorrect solutions were readily available, all 39 participants needed to re-examine search results and web pages to try and ind additional solutions (C23). For task 2, 21 (54%) of participants also tried to search for an entirely diferent type of solution (for example, some swapped from searching for how to change settings to looking for an App) (C24).

Frequent SELECT and USE tactics
We next describe the SELECT and USE tactics that more than half of the participants used. For a full list of SELECT and USE tactics see Table 7, Appendix A.
23 (59%) participants selected and used information objects for their visual clarity (SU1). For instruction type tasks, pictures and lists were considered particularly clear and easy to use. 37 (95%) participants described selecting and using information objects because it matched their information requirements in some way. 27 (69%) participants selected and used information objects because they were broadly relevant to the topic (SU8). This tactic was also used to avoid information objects that were not on topic. 21 (54%) of participants described selecting and using exactly what they were looking for (SU9). This tactic was used to re-ind information on the phone that had been found on the computer. Visual cues, such as icons, were often used to help relocation. The most frequently used match tactic was to use related information (SU11). 28 (72%) participants had to use this tactic because despite making their queries speciic (C9) to the phone provided, the search system returned results for other phone models. For many this was deeply frustrating. As well, 22 (56%) participants described needing to select and use alternative solutions (SU28). This was particularly employed when participants were unable to resolve task 2 with the solution frequently presented in the search results. Use of this tactic is indicative of the complexity of the task and also of participants trying hard and not giving up.

Frequent MANAGE task tactics
We next describe the MANAGE tactics that more than half of the participants used. For a full list of MANAGE tactics see Table 8, Appendix A.
Because the task is externally assigned, to ensure that they were on track (M1), all 39 participants checked that the solutions they found matched the requirements in the task statement.21 (54%) participants considered the likely solution prior to searching (M7). We discuss this further under 4.2.4 Knowledge use. As the tasks were all based on the same broad topic (phone safety), 22 (56%) participants described searching for all three tasks simultaneously (M10). One participant described searching for a single solution that could resolve all three tasks.
All 39 participants decomposed the tasks into smaller sub-tasks (M22). The two most frequent sub-tasks were (1) to ind information about the phone (model, operating system, version of operating system etc.) because we did not supply all the information they needed in the task statements. (2) To ind the location of menu items and apps on the phone. In part, participants needed to do this because they were swapping between devices and also because many of the search results included solutions for other phones with diferent menu structures.
Although 4 (10%) participants described taking a break and swapping to diferent tasks (M12), as this study had to be completed in a single search episode, participants could not take a break and return a diferent time. This was particularly problematic for one participant who feeling frustrated would rather have taken a complete break and come back to the task later.

Knowledge use
17 (44%) participants reported as having technical knowledge and 30 (77%) as skilled searchers. All our participants owned smart phones. 11 (28%) owned an android device but 13 (33%) of our participants had never used an Android operating system (operating system of supplied phone). Only one participant reported that they had previously tried to change phone safety settings, but 24 (62%) had read about or discussed phone restrictions for child safety. We could ind no numerical relationships between having any of these types of knowledge and search tactics used. Participants did describe though how technical knowledge and task familiarity inluenced their tactic use. For some tactics, knowledge is integral to the tactic and so we report frequency of use for this type of tactic (e.g. SU27 INTERNAL VERIFICATION: Comparing information sources with what is already known). For other tactics, participants describe using this tactic because they have or do not have domain knowledge but there could be other reasons for using this tactic so we provide illustrative quotes only (e.g. SU8 TOPICAL: Selecting and using information that broadly matches the task topic).
Surprisingly only 3 (8%) participants described inding terms from their own knowledge for query formulation (C2). However, it may be that participants did not verbalise this rather than it did not happen. Participants did describe how lack of knowledge meant that they looked for terms in the results and web pages that they could then use to formulate queries (C3).

"I just don't know how to start the phrase to search. … Because maybe I don't know the terms for this kind of thing so I am not familiar with it. So I need to try a lot of times and figure out. I will also see the website and see the keywords people usually use." (P38)
They also used autocomplete and other search system query suggestions (C4): "Sometimes it gives me more accurate things because sometimes I don't know how to type the correct things on the search engine, and this help me get more close to what I want to see" (P27).
Knowledge inluenced the use of SELECT and USE tactics. Participants who did not have domain knowledge described selecting results that broadly matched the topic (SU8) -"Just to get a situation awareness of what is going on rather than trying to do one speciic problem." (P17). Although information objects were selected for their visual clarity (SU1) because these were instruction type tasks, they were also selected by those who were unfamiliar with the phone because "I really like the pictures because it is something at least for me who know nothing about the technology, they are very easy" (P26).
Those who self-reported as having high technical knowledge described selecting up-to-date information (SU15), as they realised that technology changes and thus the solution would change too. "I am seeing pages that are old and probably not relevant to this version of the operating system" (P14). They selected results based on rank position (SU22), because, in their experience, Google's ranking is good for this domain and they therefore trusted the order of the results. "What I think is the irst result always could give me what I want. So just go to the irst one" (P20). They also used their knowledge to select the best solution (SU3). "So ideally things like parental control you want to have built into the operating system itself" (P19). Those with self-reported low tech knowledge described applying a cost/beneit to selecting information. They selected solutions because they thought it would be quick (SU5): "I thought it would be a quicker ix than downloading an app but I couldn't do it so I spent a while looking at the phone trying to …" (P24). Solutions were also selected because they were easier to understand (SU7). Seeing code and technical jargon was often particularly of-putting for those with low tech knowledge, who were concerned that the solutions would be too diicult for them to implement.

(28%) participants selected information because the solutions veriied what they thought they already knew (SU27). "I think it
can be done because I have a Samsung tablet and I have two profiles." (P13). This was not always a successful strategy because the solution for task 2 was diferent for the phone provided in this study. 22 (56%) participants also described using their own knowledge (SU14) to resolve sub-tasks. For part of task 2, participants needed to select the correct PEGI rating (www.pegi.info/page/pegi-ageratings). 20 (out of 22) based the rating on what they thought was best, rather than the one that best matched the task requirements. This is not simply a matter of least efort as many of the participants went on to extend task 2 (M14) and restrict the ilms and books that could be downloaded too.
"Under 5 because I was thinking about my niece." (P25) "The age of the child here wasn't mentioned so I just go for 7 years old because it seemed to be quite relevant with the description. I tried to sort out violence with videos and any other source related to that. Bad language. So I just assumed it was for someone who was already able to read. So not three years old. 12 it matters less because we know how they speak at this age. I just deduce 7 I guess." (P23) Knowledge particularly inluenced the MANAGE task tactics. Logically task 2 should be completed before task 1. Only 6 (15%) participants considered the logic of the task (M9), and they did this because they were already familiar with phone proiles.
"So I went directly to look for the restricted options. Because I already knew that with a restricted profile you can do all the task over here and maybe answer this question because maybe the restricted profile have other things more than just restrict apps." (P13)

"Because if you can set up a profile then you can select what can be accessed" (P12)
21 (54%) participants thought about likely solutions prior to searching (M7). This tactic worked well for those with technical knowledge but less well for those without. Those with a good understanding of phone technology looked for a solution that is central to the phone and rejected apps as a solution. Some of those less technically oriented started by looking for an app and never considered other solutions.

"Because I don't think there is an option to restrict content from the phone. I'm not sure about that I should have searched in the beginning is there an option in the settings. I just thought straightaway that I needed an app." (P18)
3 (8%) participants changed the task to it what they already knew on this topic or what they had found (M19). For example, P21 knows that there are dangers in YouTube and adapted the task to this.
"Because in my family we have a nephew and when he was younger we used to give him a tab so he always watched YouTube. But suddenly we started seeing that he was watching some video games that are violent and we think it is a dangerous thing and we should try to control it and we told that to his family. To his parents, sorry. That is why I always link it to YouTube." (P21) Participants unfamiliar with the task needed to further decompose tasks (M22) than those who were familiar as they also needed to search for the meaning of terms, concepts and objects, and ind the settings menu on the phone. For some participants this was a crucial stage.
"It was about telling me, go to the setting and then user on how to control the app downloading. But the problem was, yeah from here I was trying to go to the user setting. I couldn't find it quite easily. I am not familiar with Android." (P25) Those unfamiliar with the task were aware of the need to adapt tactics to what for them is an unfamiliar domain (M17). 22 (56%) participants relected on their familiarity with aspects of the task and how this inhibited performance (M24).

"It is the first time I'm using Android. I'm not familiar with it." (P12)
"It was hard because I've never had to be in the mindset of child safety thing because it was a bit foreign to me like. I've never had to think about making my phone child safe. I actually didn't know where to start with it. Which I felt bad about. Yeah. I should know." (P35)

People had a lot of tactics in their repertoire, allowing them to cope with lack of success
All participants had a large range of tactics in their repertoire and there was considerable diversity in tactic repertoires across participants. We could not identify any relationship between self-reported search skill and tactic use.
What we saw was that our participants were tenacious searchers and this meant that they used a wide range of tactics. When solutions did not work, they did not give up, they kept on searching, and kept on changing tactics. We think there are three reasons for this. Firstly, the tasks were designed to resonate with the participants. People recognised themselves in the task either as the friend or the person who would ask the problem. Many of the participants seemed genuinely interested in the topic of the task, and indeed when we asked participants to stop searching one participant did not want to inish, a second carried on searching using their own mobile phone and a third asked if we could email a list of websites on the topic. Secondly, our participants were challenged. They wanted to resolve the tasks and were frustrated when they could not. Thirdly, our participants did not believe that there could be no solution to task 2, and so carried on searching. While it could be that some participants assumed that we would not set a task that was not achievable, the reason for persisting given by our participants was that they feel protective towards children. They could not believe that it was not possible to set up proiles on the phone provided and considered the phone company irresponsible.

Universal tactics but application is task specific
Our study supports the idea of a universal set of tactics (Savolainen, 2017), as many of the tactics for the how-to task in this study have been seen in other studies; most notably those identiied by Bates (1979b) and Barry and Schamber (1998). However, some tactics common to other studies were either not described or only infrequently described in this study. This is likely linked to the type of task and a limitation of our study design.
An example of a tactic not seen in our study that is linked to the type of task, is accuracy of information seen in both (Barry & Schamber, 1998;Savolainen & Kari, 2006) studies. As our tasks were instruction-based, participants were not concerned with accuracy while selecting information, instead they looked for good solutions and then tested the accuracy of information when implementing solutions on the phone. Similarly, the most commonly used selection criteria in this study were selecting/using exactly what is looked for (SU9) and selecting/using information that is diferent from but related to what is looked for (SU11), whereas in Savolainen and Kari's (2006) study speciicity, topicality, familiarity, and variety were used most frequently. We suggest that this shows the inluence of task and information retrieval system performance: in our study the search results were poor and participants had to select related information, furthermore because they were using two devices they needed to search for exactly the same information again. In Savolainen and Kari's (2006) study, participants are searching for self-development, these are likely open-ended topics which may have been searched before and which will have a range of appropriate information.

Frequently applied tactics linked to the task and search system response
Prior to conducting this study, we had anticipated that tactic use would be particular to the type of task because information must be used to implement solutions on the phone, and a task would only be successful if the information searched for could resolve the information problem. In Table 4, we summarise the reasons participants gave for the frequently used tactics.
Four of the reasons given are related to the task: it is assigned, it is instruction-based, the solution must be implemented on another device, the task is complex. Although the assignment is a product of a lab-based study, search is social (Evans & Chi, 2010) and when an obvious answer is not available, many people turn to 'techy friends' to help ix a computer problem. Participants used the information given to them "by their friend" to formulate queries (C1) but because crucial information was missing they also needed to ind additional information (C5). They also needed to check that the solutions they were implementing matched the task requirements (M1). That the tasks had an instruction information use that needed to be applied to a diferent device inluenced tactic selection. Our participants needed to ind solutions that were speciic to the phone model and operating system (C9), locate items on the phone (C10, C15, C16, SU9) and they wanted to use information eiciently (C19, SU1). Four tactics (C17, C23, SU28, M22) are likely linked to the complexity of the task. Search task complexity is based on three dimensions: number of subtasks, number of facets and indeterminability (Wildemuth, Freund, & Toms, 2014). It is primarily the indeterminability of our task that makes it complex. The complexity is possibly further compounded because our participants did not expect the tasks to be hard. To deal with the indeterminability our participants needed to keep their options open (C17) and ind alternative solutions (C12, C23, SU28). Our participants also need to decompose the tasks (M22) into a number of subtasks; a second aspect of complexity. They needed to do this because we did not give them enough information in the task statement and they had to implement the solution on another device that they were also unfamiliar with. That the search system returned results that were general rather than speciic to the task accounts for three frequently occurring tactics. Participants reformulated their queries by changing terms (C12) or approach (C24). They used related information (SU11) when they could not ind exactly what they were looking for. Searcher persistence (noted at the start of this section) accounts for why people kept using these tactics when earlier approaches were unsuccessful (see Section 5.1).
Two tactics are best described as "short cuts". To reduce searching participants tried to combine the tasks (M10), and if participants could avoid searching for information by using what they thought were likely solutions (M7), they did.
Selecting information that broadly matches the task topic (SU8) is linked to many participants working in what is for then an unfamiliar domain, either because the task is unfamiliar or they do not have much technical experience. We discuss this further in Section 5.4 Inluence of domain knowledge.

Influence of domain knowledge
We consider how domain knowledge inluenced tactic use next. As the number of participants were too few and what constitutes domain knowledge (technical knowledge and task familiarity) too messy for us to group by frequency of use, in Table 5, we group tactics according to whether participants stated that they used this tactic either because they had or did not have domain knowledge.

Table 4
Reason for frequent tactic use. Frequent tactics (used by more than 50% of participants) Reason for use C1 EXTRACT: Finding terms in the task statement The task is assigned C5 OTHER SOURCE: Finds terms from another source (i.e. a source other than C1-C4) M1 CHECK: Checking solutions and potential solutions against the task statement C10 SPECIFY: An exact term is selected because a speciic item is searched for Two devices SU9 EXACT: Selecting exactly what was looked for C9 PINPOINT: A term is selected that restricts the results to the topic / an aspect of a topic Instruction based task C15 LOCATE TERM: Locating a known term on the interface C16 LOCATE ALTERNATIVE: Locating alternative term(s) on the interface C19 SIMULTANEOUS: Viewing information on one device while implementing solution on another SU1 VISUAL CLARITY: The presentation and format of the information makes it easy to use C12 VARY: A term is selected to replace another term to change the results Task complexity C17 RECORD: Keeping evaluated information available for later use either by opening new tabs or taking notes C23 REEXAMINE: Returning to search results and / or websites that have already been examined to ind additional information SU28 ALTERNATIVE: Finding a new solution to replace previous solution M22 SELECT: Break the task into sub-tasks and work on one sub-task at a time C12 VARY: A term is selected to replace another term to change the results Solutions are highly speciic but the search results are general SU11 STRETCH RELATED: Selecting information that is diferent but related to what was looked for C24 REROUTE: Use an entirely diferent approach to ind additional information sources SU8 TOPICAL: Selecting information that broadly matches the task topic Working in an unfamiliar domain M7 LIKELY SOLUTION: What solutions are likely "Short cuts" M10 COMBINE: Searches for all tasks at the same time Table 5 Tactic use inluenced by knowledge.

Tactics
Reason for selection SU3 BEST SOLUTION: Selecting an object because it is known to be the most efective solution Task is familiar / has technical knowledge SU15 CURRENCY: Selecting information objects that are up-to-date SU22 TRUST: Selects information at the top of a ranked list based on trust M9 LOGIC: Logically makes sense to change the task order SU27 INTERNAL VERIFICATION: Comparing information sources with what is already known C3 TRACE: Examining results page and web sites for terms Task is unfamiliar/has no technical knowledge C4 SUPPORT: Using search system functionality (autocomplete, people also ask, related links) to ind terms SU1 VISUAL CLARITY: The presentation and format of the information makes it easy to use SU5 TIME CONSTRAINTS: Selecting an information object based on time available e.g. it will be quicker SU7 ABILITY TO UNDERSTAND: Selecting an information object that can be easily understood SU8 TOPICAL: Selecting information that broadly matches the task topic M17 DOMAIN: Adapting tactics to the task domain M22 SELECT: Break the task into sub-tasks and work on one sub-task at a time SU14 OWN SOLUTION: Use own knowledge instead of searching Finds way to use own knowledge We had expected that those with high technical knowledge would have access to more search tactics when resolving a how-to technical tasks than those with little technical knowledge even if they were familiar with the task. However, participants only described four tactics that used domain knowledge, and what constitutes domain knowledge varies for the diferent tactics. Those with high technical knowledge described selecting results because they are up-to-date (SU15) as they recognised that the recentness of published information is important in the technical domain. Those experienced at searching in the technical domain also selected results at the top of a ranked list because they trusted Google's ranking for this domain (SU22). Those familiar with the task selected solutions that matched what they thought they already knew (SU27), and as Bhavnani (2001) found those who were familiar were more adept at sequencing the task (M9). Those more technically experienced were able to choose between solutions (an app vs changing the phone setting) presented in the results based on what they already know to be more efective (SU3).
Participants described 8 tactics that they used to compensate for not having domain knowledge. Lack of familiarity with the task meant participants need to ind query terms (C3, C4), gain a general understanding of the topic (SU8), and breakdown the tasks (M22). Lack of familiarity with the domain meant that participants need to ind solutions they could understand (SU7), information that was easy to use (SU1). They were also aware that they needed to adapt their usual search patterns (M17). Participants without domain knowledge were also more aware of time constraints and selected solutions because they thought they would be quick to implement. This again highlights the inluence of Saracevic's (2007) Situational and Afective forms of relevance, where results had to be ones that they could themselves use, given their technical knowledge and device experience.
We had expected that knowledge would very strongly correlate with success, but it did not. Knowledge was often not used well. Like many real-life issues, there are a variety of approaches to resolving this task, and what participants did was often coloured by what they already knew or thought they knew (M7, SU27, SU14). In general those who used an emergent strategy (develops during search) as opposed to a deliberate (existed prior to search) (Savolainen, 2016), and adapted their tactics to the situation, fared better than those who planned ahead. We also observed that participants tended to use their own knowledge (SU14) if they could, that went beyond the immediate task. Although we collected what we thought were a wide range of familiarity data relating to the task (see Appendix B), participants used knowledge that we did not anticipate. For example, they used their knowledge of children (often based on experiences of family members) and what they considered appropriate for children. This re-application of knowledge (SU14) to bridge their knowledge gap may be a good clear example of 'Create Information' from Godbold's (2006) Navigating Gap diagram.

Realistic tasks add their own kind of complexity for both searchers and researchers
Although the assigned task was not entirely achievable (in that the correct solution to task 2 is that there is no solution), no participants exhibited signs of distress at being unable to complete it. In fact, most participants were enthusiastically engaged such that some did not want to stop searching; these participants expressed that child safety was so important that the information should be clearly available, and if anything, were surprised that it was not.
We set a task that was challenging but still we were surprised by how diicult to achieve it was for many of our participants. As no technical knowledge was required to implement the solutions on the phone, our tasks were a test of search skill (rather than technical ability). However, it was not that participants found it diicult to ind solutions per se i.e. it was not diicult to ind relevant information. What participants found diicult was inding solutions that worked. That so few studies consider whether search resolves the problem that led to the use of search systems (Kelly, 2007;Wildemuth, 2004) is a concern.
Many research studies aim to test a particular hypothesis, or focus on the impact of particular variables. To achieve such aims it makes perfect sense to design laboratory studies that try to control other confounding variables. By contrast we were committed to trying to design a realistic how-to technical task in order to see what people actually do. Our indings, as outlined above, surprised us, and this was caused at least in part by a series of complexities that sneak in when you deliberately do not try to control for certain real-world complexities. In particular, we found it revealing how additional issues arise when you do not stop at the point of people inding relevant information, but watch what happens when people try to put that information to use, hit a problem and have to do additional searching.

Design implications
Search systems should be designed to respond to users' tasks (Toms, 2011). We next make suggestions for how search systems could be designed to support how-to technical tasks.

Help with specificity and key terms
In part, these tasks were diicult because people did not know what they needed to know to resolve them. Many participants felt that Google could support them more by guiding them to be more speciic in their queries. Suggestions included (1) prompts of "which device" while entering queries, (2) prompts of information to include running in a background browser tab, (3) a list of all phone models that could be iltered on the results page, and (4) support the C3 TRACE tactic better by identifying 'key domain terms' in the search engine results pages, rather than just those terms highlighted in snippets that are from the query. While these ideas have so far been technology-search speciic, the idea can easily translate into other domains like medical queries, where the technical detail of the queries really matters.

Query terms more clearly reflected in search results
Even when participants made their queries speciic to the phone model and operating system, this was often not relected in the search results. Varying key terms in queries did not diversify the results much either. Clustering similar results may give searchers conidence but for this type of task where the solution is speciic (and not general to all phones) and where there are both good and bad solutions, our participants wanted search systems that are more responsive to key terms and changes in query terms.

Providing major directions, overviews and visual information
Although participants found auto-complete helpful, the recommendations are only continuations of what has been entered, and use of this functionality did not noticeably improve task success. What participants would like is for Google to suggest alternate words that guide them towards more fruitful 'sets of terminology'. "If those give the good kind of completion, a semantic idea of: if I'm typing in 'child safety' that it would immediately potentially suggest 'parental controls' or something like that." (P14). Although search engines do typically provide query suggestions, this particular focus is more in line with the use of idea tactics for search suggestions (Kelly, 2009), to correct or direct the higher-level search approach to related or more successful pursuits. Some participants commented on the usefulness of the information box for this kind of overview guidance. An opportunity is to expand the use of information boxes to establish overviews, especially in relation to large and evolving information areas like child safety online. Participants also wanted more visual information e.g. list, videos and pictures.

Quality and recentness of results
For these search tasks, participants questioned the quality of the results in terms of authority and recentness. Forum discussion sites were often at the top of the results list. Even though lots of the most useful information was in forums, several participants wanted to be directed to sites owned by Android and the phone brand. They considered these sites more authoritative and were frustrated that often when searching for information about Android and the phone brand, these sites did not appear in the results list.
Although the recentness of published information is important in the technical domain, recentness was rarely used because while Google does incorporate the date of publication in snippets, many of the results did not have this information.

Methodological issues and limitations
This was a study of an assigned task conducted in a laboratory setting, and this had some inluence on the tactics used. To make the assignment realistic we simulated an assigned task "your friend has asked you […]". Several participants commented on the ecological validity of the task, particularly those who self-reported as being technically competent. For others, the task was also ecologically valid but they considered themselves "the friend". Nevertheless, our study did have ecological validity weaknesses. To manage our study, we set a time limit to complete the tasks whereas in real-life people can carry on searching till they resolve the problem. When we asked our participants if they had enough time, almost half felt they didn't. The limitations that participants told us unprompted was that they would like to have taken a break and come back later, and that all searching had to be done online. That was a problem for many, as participants noted that under normal circumstances they would also consult an expert or speak to family and friends.
Our approach to identifying tactics was reliant on participant descriptions. Although this provides the deep and rich insights into tactic use to meet our objectives, there are also limitations. Firstly, participants can inadvertently overlook details, particularly those that are routine and that for the participant are obvious. For example, only 35 (90%) participants describe how they found terms to use in query formulation, yet they all clearly found terms from somewhere. Secondly, there are so many tactics that it is diicult for participants to describe everything. Our reported results are the superset of tactics discussed by all participants, but we cannot, using this approach, either objectively log or comprehensively identify every tactic used, when so many involve internalised actions. Nor can we report how frequently each participants' used each tactic. While we report the frequency of occurrence of tactics across the dataset (i.e. how many participants used this tactic), some caution is required in interpretation and so we do not test for statistical signiicance.
Our study aimed to try and understand more about how people dealt with everyday how-to technical tasks. We found that, for our participant sample and the tasks given, people used a substantial number of diferent search tactics. Our participants were well educated and comfortable with technologies, even though they did not all identify as experts. This may be representative of a growing percentage of the wider population, but there is always the risk of a sampling bias. Future work could check if it is indeed widespread in many settings.
We did not rotate tasks, and participants were free to resolve the tasks in any order. This meant that we could see how our participants' tackled the whole work task. However, because participant fatigue meant that they tended to describe tactics the irst time they occur (rather than each time) it was not possible to distinguish tactic use between the three tasks in our analysis.
We were unable to identify any numerical diferences between use of tactics and task familiarity and technical knowledge. We argue that this is because what constitutes domain knowledge for this type of task is complex. However, our tactic identiication is ine-grained (77 tactics); such schemes are not suitable for statistical analysis (Wildemuth, 2004). This also means that our scheme will be of limited value to those wishing to analyse sequences of tactics. However, the large list of tactics that we and others have identiied is an important reminder of the complexity of searching, particularly in open-ended more realistic tasks as done by people rather familiar with online searching. We believe that the rich and deep insights that ine-grained schemes aford are useful for system designers when considering the functionality that search systems should ofer and the sheer variety of tactics that people may employ while using such systems. Optimising for a small number of tactics is unlikely to be useful.

Future work
Only 23% of participants successfully completed all three tasks. The success rate was much lower than we had anticipated. Whether we had designed a task that was unusually diicult or whether people do ind these types of tasks diicult is unclear. We suspect both are true -some how-to tasks will be simple but others more complex. In future work, we intend to investigate tactic use for how-to tasks with varying complexity. Our how-to task was in the technical domain. We could expect that the domain and not just the type of task inluenced tactic selection (Wildemuth, Kelly, Boettcher, Moore, & Dimitrova, 2018). And indeed in our study, those familiar with the domain selected some tactics (e.g. SU15 CURRENCY) because of they were thought particularly applicable for this domain. In future work we would like to compare tactics for how-to tasks in diferent domains.
In our study, we only considered a search successful if the information found could be used to resolve the information problem (i.e. participants needed to be able to implement the solutions on the phone provided). For an everyday how-to task, we could do this because (a) what counts as successful resolution is more deinite for how-to tasks (i.e. typically it works or it doesn't) whereas for example with research tasks what counts as successful resolution is more of a judgement call (b) implementation did not require any technical skill -search skill was need because inding the correct solution was diicult. This meant we could interpret successful implementation as "search success". In future work we intend to investigate how success can be measured for other types of task and for tasks where the solutions are diicult to implement.
Technical how-to tasks (especially when the searcher fails to ind the desired result in a single shot interaction) have many parallels with Search as Learning (Hansen & Rieh, 2016). There is an ongoing process of assessment and discovery, and ideally some relection on what is an is not working. We want to further investigate these parallels by e.g. including measures of learning over time.

Conclusions
Although most search tasks may be accomplished by a single query and selection of the most relevant result, there are many reallife tasks that are somewhat more complex, requiring iterated and integrated exploratory searching. It used to be that only information professionals were experienced searchers, and in the early days of the web they had more search tactics than the general population. Nowadays a lot of people are experienced web users (Thatcher, 2008) and our participants had a lot of tactics in their repertoire regardless of technical knowledge. However, that does not mean that inding easy to understand how-to solutions to technical problems is now rendered trivial. Certain tasks, like ours, remain challenging, even for people with considerable domain knowledge and a wide repertoire of search tactics. Indeed, only 23% of our participants were able to successfully complete the whole task.
We conclude that having a range of tactics, although highly desirable, does not on its own ensure task success; it is knowing when and how best to employ these tactics for the task at hand that makes a diference. Domain knowledge afected the choice of tactics used (although not necessarily towards better task success), and the most frequently used tactics were linked to the type of task and system response. In our study, searching also required the assessment of the applicability of results to the underlying work task at hand. One of our key conclusions is that this diference between inding information and being able to use it afected activity including relevance judgements and, therefore, the tactics used. It was not hard for participants to ind information, what was hard was for participants to ind information they could use. It is important to continue to develop our understanding of these more complex information needs in order to develop resources and advice on how to address them.
Select and use tactics involved those that participants use to select objects; that is selecting which search results to visit, and what information is used in problem resolution. This group of tactics (Table 7) is used when examining results, and also when implementing solutions on the phone.

Table 6
Tactics used to control the low of the search; counts in the right-hand columns show number of participants rather than number of times.

CONTROL tactics
No. of participants (n = 39) Finding terms to use in query formulation 35 C1 EXTRACT 1 : Finding terms in the task statement 24 (62%) C2 OWN KNOWLEDGE: Finding terms from own knowledge 3 (8%) C3 TRACE 2 : Examining results page and web sites for terms 14 (36%) C4 SUPPORT: Using search systemfunctionality (autocomplete, people also ask, related links) to ind terms 13 (33%) C5 OTHER SOURCE: Finds terms from another source (i.e. a source other than C1-C4) 28 (72%) Selecting which of all available terms to use in query formulation 39 C6 REDUCE 2 : Selects only some of all available terms 7 (18%) C7 EXHAUST 2 : Selects many or all available terms 5 (13%) C8 COMPROMISE: Selects the terms that are available even if they are not the best terms 2 (5%) C9 PINPOINT 2 : A term is selected that restricts the results to the topic / an aspect of a topic 32 (82%) C10 SPECIFY 2 : An exact term is selected because a speciic item is searched for 25 (64%) C11 PARALLEL 2 : A term is selected to broaden the range of results 15 (38%) C12 VARY 2 : A term is selected to replace another term to change the results 20 (51%) C13 CORRECT 2 : A term or terms are selected to correct errors in query formulation 13 (33%) C14 INFO TYPE: A term is selected to restrict results to a particular answer type, format, media or domain.
14 (36%) Locating terms on an interface (e.g. computer / phone) 34 C15 LOCATE TERM: Locating a known term on the interface 33 (85%) C16 LOCATE ALTERNATIVE: Locating alternative term(s) on the interface 27 (69%) Maintaining information sources 39 C17 RECORD 2 : Keeping evaluated information available for later use either by opening new tabs or taking notes 30 (77%) C18 PARALLEL HUB & SPOKE 3 : Opening multiple results in new tabs and thereby making unevaluated information available for later use 15 (38%) C19 SIMULTANEOUS: Viewing information on one device while implementing solution on another. If information and implementation on same device then notes used.

(100%)
C20 CLOSING: Closing down recorded information sources to make information space more eicient 5 (13%) Reinding information sources 17 C21 RUN AGAIN: Repeating a process exactly as was to double check 12 (31%) C22 FIND AGAIN: Repeating a process to reind information 9 (23%) Finding additional information sources 39 C23 REEXAMINE: Returning to search results and / or websites that have already been examined to ind additional information 39 (100%) C24 REROUTE: Use an entirely diferent approach to ind addittional information sources 21 (54%)

Table 7
Tactics used to select results and use information to solve the task; counts in the right-hand columns show number of participants rather than number of times.
SELECT and USE tactics No. of participants (n = 39) Clarity of information 23 SU1 VISUAL CLARITY 4 : Selecting and using an object because the presentation and formatmakes it easy to use 23 (59%) Efectiveness 5 of solution 30 SU2 A SOLUTION: Selecting an object because it provides a solution 18 (46%) SU3 BEST SOLUTION: Selecting an object because it is known to be the most efective solution 13 (33%) SU4 ACCOMMODATION SOLUTION: Selecting an object because it is a good solution given the situtation 6 (15%) Accessibility 5 of information -applies a cost beneit (weigh 2 ) to selection and use 22 SU5 TIME CONSTRAINTS 5 : Selecting and using an information object based on time available e.g. it will be quicker 14 (36%) SU6 AFFORDABILITY 4, 5 : Selecting and using an information object based on inancial cost 4(10%) SU7 ABILITY TO UNDERSTAND 5 : Selecting and using an information object that can be easily understood 16 (41%) Selecting and using an information object because it matches information requirements 37 SU8 TOPICAL 5 : Selecting and using information that broadly matches the task topic 27 (69%) SU9 EXACT: Selecting and using exactly what was looked for 21 (54%) SU10 CONTRARY 2 : Selecting and using information that is opposite to what was looked for 4 (10%) SU11 STRETCH 2 RELATED: Selecting and using information that is diferent from but related to what was looked for 28 (72%) Selecting and using information because it is available 5 26 SU12 BEST: Selecting and using the best solution that is available in the results (but there may be better solutions) 3 (8%) SU13 OFFERED: Selecting and using what is ofered (without evaluating irst) 3 (8%) SU14 OWN SOLUTION: Selecting and using own knowledge instead of searching 22 (56%) (continued on next page) Selecting and using information based on time 5 6 SU15 CURRENCY 4, 5 : Selecting and using information objects that are up-to-date 6 (15%) Selecting and using new information 10 SU16 NOVELTY 5 : Selecting and using information objects because they are new to the user. Also avoiding objects that are not new 9 (23%) SU17 FAMILIARISE: Selecting and using a range of information sources to get familiar in unfamiliar landscape 5 (13%) Selecting and using information based on its qualiity 5 23 SU18 REPUTATION 5 KNOWN: Selecting and using information from a known resource 7 (18%) SU19 REPUTATION 5 OFFICIAL: Selecting and using information because it is from oicial organisations 11 (28%) SU20 TRIAL: Selecting and using any information, and then testing for quality 13 (33%) SU21 GENERAL PUBLIC: Selecting and using information posted by others having similar experience 10 (26%) Selecting and using information based on position 11 SU22 TRUST: Selecting and using information at the top of a ranked list based on trust 9 (23%) SU23 ORDER: Selecting and using information according to position in ordered list based on own rules 2 (5%) Selecting and using information to verify 6 28 SU24 DISCRIMINATE: Comparing information sources to determine which one is better 6 (15%) SU25 CONSENSUS 4 : Comparing information across multiple sources so as to determine a consensus 12 (

Table 8
Tactics used to manage the overall resolution of the task; counts in the right-hand columns show number of participants rather than number of times.