HCI Models, Theories and Frameworks: Toward a Multidisciplinary Science
by John Carroll

Carroll, J. M. (Ed.) (2003). HCI Models, Theories and Frameworks: Toward a Multidisciplinary Science. San Francisco: Morgan Kaufmann publishers.   ISBN: 1-55860-808-7.
This collection of tutorial articles is an appropriate survey for the graduate level student. [DS]

Jump To:   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15

Ch. 1:  Introduction
Ch. 2:  Design as Applied Perception (visual senses)
Ch. 3:  Predictive vs. Descriptive Models, Fitt's Law
Ch. 4:  GOMS, KLM, Advanced / Modified GOMS
Ch. 5:  Cognitive Dimensions of Notations Framework
Ch. 6:  Users' Mental Models
Ch. 7:  Information Foraging Theory, Optimal Foraging Theory, Scatter/Gather (evolutionary)
Ch. 8:  Collaborative Technologies / Distributed Cognition
Ch. 9:  Cognitive Work Analysis (CWA)
Ch. 10:  Clark's Common Ground Theory (as used in CMC)
Ch. 11:  Activity Theory
Ch. 12:  CSCW Research (Psychological foundations)
Ch. 13:  Ethnography, Situated Action, Ethnomethodology
Ch. 14:  Computational Formalisms
Ch. 15:  Design Rationale as a Theory

Chapter 1:  Introduction

HCI lies between social and behavioral sciences; computer and information technologies.  It is the fastest growing CS field, and needs a multidisciplinary approach to keep flourishing.

HCI practitioners:  analyze and design user interfaces, integrate technology to support human activity

• in the past:  WIMP interface, voice / video interfaces

• today:  better input / display devices (mobile computing), information visualization (digital libraries), navigation techniques (virtual environments)

Initially, HCI brought cognitive science theories to focus on software development

Activity theory brought into HCI (Marxist foundations), complemented cog. sci.

While a multidisciplinary approach helps HCI theories, a tradeoff is the fragmentation of HCI (it's very difficult to make sense of the vast, diverse science of HCI; to synthesize a comprehensive, coherent methodological framework)

Goal of the book:  To 'survey' the many approaches to HCI; compare / contrast them
 - each chapter is a different approach / method

(top)

Chapter 2:  Design as Applied Perception

Much of human intelligence can be characterized as our ability to recognize patterns

Vision dominates as the primary sense:  engaging 50% of the cortex and more than 70% of all our sensory receptors

Why we care:  we have a fundamental assumption that there is such an entity as the human visual system
 

display design - map data into a visual form that is matched to our perceptual capabilities

• the same perceptual mechanisms that enable us to perceive the world enable us to perceive information patterns in computer displays

In Cognition, humans share a similar muscular / skeletal structure, but have a highly adaptable neural structure, which allows for a high degree of adaptation (humans are highly adaptive machines)

Information Psychophysics - concerned with how we can see simple elementary information patterns (paths in graphs, clusters of items, correlated variables, etc.)

• Theory of Affordances (J. J. Gibson) - a highly relevant theory, saying that affordances are physical properties of the environment that are directly perceived (before HCI / computers)
   
• Since a computer screen doesn't have a physical environment, we use a metaphor - which says that a good user interface is similar to a real-world situation
   
• Information Psychophysics can be seen as a component of a larger set of cognitive models, such as ACT-R (Adaptive Control of Thought - Rational [Anderson]) and EPIC (Executive-Process/Interactive Control [Kieras]) which focus on the cognitive processing of sensory information

 

3 Stage Model of the Visual System:  this happens very fast - our brain's process when we see stuff

• Stage 1:  Early Vision:  our visual image is analyzed into color, motion and shape

  • involves eye optics, photoreceptors
  • use trichromatic system to see colors- segment into Red / Green / Blue
  • use luminance to see detail / contrast
  • the theory of color vision is that we separate color into red-green, blue-yellow, and black-white opponent color channels
  • Opponent colors explain why red, green, yellow, blue, white and black are special colors in all societies, and why we should use these colors first if we need to color-code things
  • Since these colors can't convey patterns, we use luminance contrast to show detail
     
  • preattentive processing theory - uses early stage processing, and tells us how to design things so that they POP-OUT of a cluttered display.  (Color, motion, texture, stereoscopic depth can be preattentively perceived, and understanding these limitations is critical in making displays that can be rapidly interpreted)
     

• Stage 2:  Pattern Perception:  we split what we see into segments:  a 2-D image, pick up edges, etc.

  • we can display data using patterns that are easy to perceive, which facilitates problem solving

  • GESTALT psychologists (Wesheimer, Koffka, Kohler) first theorized of pattern perception:

    • Proximity - entities that are close together are perceptually grouped
    • Good Continuity - smooth continuous lines are more readily perceived
    • Symmetry - symmetric objects are more readily perceived
    • Similarity - similar objects are perceptually grouped
    • Common Fate - objects that move together are perceptually grouped
    • Common Region - (added later) - objects in enclosed spaces are grouped
    • Connectedness - (added later) - objects connected by continuous contours are perceived as related
       

  • In HCI, visual displays can support creative thinking and problem solving
     

• Stage 3:  we perceive objects (infer based on stages 1 & 2) out of what we see

  • working memory - core of modern models of cognitive processing

    • there are several components of working memory:  visual, verbal, etc...

  • Structured object perception (Biederman) - we perceive objects as composed of simple linked 3-D solid shape primitives, called GEONs.  There is considerable difference between high-level (object perception, color is secondary) and low-level vision (color perception, etc)

In the above system, feedback loops can modify what we see.  Higher stages can feedback to lower stages to modify what we see.  Lower stages are more robust, while higher stages involve more inference making.

In the above system, culture makes it difficult to design by visual patterning since aspects of displays owe their value to cultural factors (R, G, Y, B are different colors in every culture).  Some cultural aspects are so hard-wired that there's no way to design around them.

(top)

Chapter 3:  Motor Behavior Models for HCI

Why model motor behaviors?  Because we need to match movement limits, capabilities, and potentials of humans - to input devices and interaction techniques - on computing systems

A model is a simplification of reality that we can use to design, evaluate and provide a basis of understanding a complex behavior of complex artifacts. 

Models lie somewhere in a continuum:    analogy / metaphor  <-- ? --> mathematical equations

• Predictive Models:  on the mathematical end
  (a.k.a. Engineering Models or Performance Models, such as Fitt's Law)

• Descriptive Models:  on the metaphorical end of the spectrum
  (Descriptive, such as Guiard's model - provide a framework / context to think about a problem)

More on Predictive Models:

allow human performance to be analyzed analytically, avoiding time-consuming experiments

predictions are a-priori, allowing for hypothetical exploration of a design scenario

Hick-Hyman Law:  an example predictive model, predicts human response time by an equation

Fitt's Law:  a predictive model for human movement (measures accuracy and amplitude of human movement; computes an index of performance to compare efficiency of different devices / interfaces)

• drawn from Shannon's theory - uses electronic noise (accuracy) and electronic signal (amplitude) to quantify a movement task's index of difficulty (ID) and predict movement time (MT) to complete a task
   
• ID / MT = Index of Performance (IP); now known as Throughput (TP)
   
• The goal of Fitt's law:  determine which devices / interaction techniques are most efficient by comparing performance measures

  An example:  keystroking on mobile phones:  ones with multi-tap vs. ones with one-key disambiguation (guesses most common word based on key presses):  second is faster based on Fitt's law, although more error prone

• Fitt's law isn't appropriate for complex tasks, like measuring typing speed on a qwerty keyboard (complex task with 2 hands)

Keystroke-Level Model (KLM):  a predictive model for predicting time to do a task by dividing into sub-tasks:

K = Keystroking
P = Pointing
H = Homing
D = Drawing
M = plus 1 Mental operator
R = plus 1 system Response operator

T_Executive = t_K + t_P + t_H + t_D + t_M – t_R

Other Predictive Models:  GOMS (see Ch. 4), Programmable User Model (PUM:  Young, Green, Simon 1989)

_{More on Descriptive Models:}

Descriptive Models provide a framework / context for thinking about a problem

Example:  Key-Action Model (KAM):  divides keys on a keyboard into 3 categories:  symbol keys, modifier keys, and executive keys (simple model, organizes keys into categories)
 

Three-State Model of Graphical Input (Buxton):  Descriptive model, says computer pointing devices follow a state transition diagram of 3 states:  out of range, tracking, and dragging (good for analyzing different types of pointing devices [touchpad, trackpad, mouse, etc]) - resulted in redesign of touchpad's to be pressure sensitive, so drag state could be induced easily
 

Guiard's Model of Bimanual Skill:  studies between-hand division of labor in everyday tasks - see chart of descriptive model on p. 41

• Features of non-preferred hand:  leads the preferred hand, sets spatial frame of reference for preferred hand, performs coarse movements
   
• Features of preferred hand:  follows non-preferred hand, works within spatial frame of reference (set by non-preferred hand), performs fine movements

Buxton & Myers found that people prefer to use two hands when not instructed, and this resulted in lower times in task completion.  This had little insight into hand use, but was one of the first papers in HCI.
 

Another study (Gibson) of keyboards found that they are right-hand biased (power keys on right side), but the Desktop as a whole is left-handed bias:  lefties can cash in on a time savings, because having the mouse to the left of the keyboard allows for the right hand to use the keypad / power keys WHILE the left hand operated the mouse (don't have to stop, switch between the two)

(top)

Chapter 4:  Information Processing and Skilled Behavior

GOMS is the primary focus of this chapter (Goals, Operators, Methods, and Selection rules:  a way to describe a task and calculate the time to complete it)

• Goals:  The users' goals:  what does the user want to accomplish with the software, and by when?
• Operators:  The actions that the software allows the user to take.  Generally it is a command such as a button press, menu selection, or a direct-manipulation action.
• Methods:  the sequence of sub-goals and operators that can be used to accomplish a goal:  such as the sequence of moves necessary to cut and paste text to move it from one location to another.
• Selection Rules:  Giving the user the personal "choice" (selection) to accomplish the goal in those times when there is more than one method available to accomplish the goal or certain task

GOMS applies to situations in which users will be expected to perform tasks that they already have mastered.  It does not work for tasks being done by novices that are trying out a new interface design.  The knowledge gathered by GOMS reflects what a skilled person (expert) will do in a seemingly unpredictable situation.

GOMS works for single-user, active systems where the system changes in unexpected ways or other people -participate in accomplishing the task.  It has been shown to work very well in analyzing user-paced, passive systems.

GOMS can be used quantitatively and qualitatively. 

• Quantitatively:  gives good predictions of performance time and relative learning time
• Qualitatively:  can help design training programs, help systems and inefficient systems following an analysis of the results of study

GOMS as a task analysis technique:  stemming from conceptual frameworks on human information processing (Carroll, p. 65):

Conceptual frameworks - informally stated assumptions about the structure of human cognition

serial stage models - we process information in a serial sequence
parallel multiprocessor stage models - we process information like parallel computing

Computational Cognitive Architectures - also called unified theories of cognition - are proposals to make cognition explicit enough to run as a computer program

The Model Human Processor (Card et. al, 1983):  ties cognitive science and engineering models:  represents human cognition as a parallel computer.  The parallel processing being done:

• Cognitive (thinking / thought process)

• Perceptual (seeing / what we observe)

• Motor (doing / what physical movements)

In the diagram above, the bottom is more theoretical while the top is more practical.  GOMS is a way designed to make human information processing something that can be measured practically.

Task Analysis Techniques - these map out the system in terms of goals, operators, methods and selection rules.  There are 3 restrictions to what GOMS models can be used for:

1. there must be a way to do the task in question

2. the task must be able to be 'routinely done' by a skilled expert

3. the GOMS analyst must start with a list of top-level tasks or user goals (provided from external sources outside of GOMS)

GOMS Models:

1. KLM:  Keystroke Level Model (Card, et. al)

   • this is the simplest GOMS:  turns a simple action into a sequence of steps for analysis

   • Time of execution = T(execute) = Tk + Tp + Td + Tm + Tr
     Operators:  K (keystroke); P (pointing); H (homing); D (drawing); M (mental preparation); R (system response time)

   • The main advantage is this allows for a very quick estimate of execution time with very little theoretical / conceptual baggage.  It is the most practical GOMS technique.
      

2. CMN-GOMS:  Card-Moran-Newell GOMS

   • named this to differentiate itself from other GOMS models as they began to appear

   • based on the serial-stage model (see diagram above), not the Model Human Processor (MHP)

   • slightly more specified than general GOMS

   • Quantitatively:  predicts the operator sequence and execution time

   • Qualitatively:  focuses attention on methods to accomplish goals.  Similar methods are easy to see.  Unusually short or long methods jump out and can spur design ideas.
      

   • While the KLM has no explicit goals, CMN-GOMS does.  The output of CMN-GOMS is in program form, so it is executable.
      

3. CPM-GOMS:  Cognitive, Perceptual, Motor GOMS (Bonnie John)

   • based on the MHP (Model Human Processor) - Cognitive, Perceptual and Motor Operators can run in PARALLEL (subject to resource and information dependencies)

   • based on parallel multiprocessor stage model of human information processing

   • CPM also stands for Critical Path Method, since the critical path in a schedule (of parallel processes) determines the overall execution time.

   • Quantitatively:  predictions of performance times can be read off the chart of CPM-GOMS

   • Qualitatively:  analysis of what portions of a design lead to aspects of the performance are easy once the models are built

   • as with KLM, selection rules are not explicitly represented in the chart because the chart is just a trace of the predicted behavior

   • This is the only model representing parallelism, so several goals can be achieved at one time, which is characteristic of expert performance of a task

Human Information Processing

• this approach can be used to model more complex human behaviors like problem solving, learning and group interaction, all of which are critical to designing complex systems

• some HCI practitioners think of GOMS as a good tool, many think it is not (too cumbersome / time consuming)

(top)

Chapter 5:  Notational Systems- The Cognitive Dimensions of Notations Framework

http://homepage.ntlworld.com/greenery/workStuff/Papers/introCogDims/index.html

Think of the point of view of the system creators (practice side, not so theoretical). 

System designers try to incorporate new HCI Theories into their designs, but have learned that some are more practical than others.  They want to use theoretical knowledge, but prefer things like checklists which often aren't too theoretical.  The authors believe that it will not be possible to deal with these new notational systems by creating new checklists.

Especially with user testing- many of the methods that have been suggested such as user testing (laborious, expensive, in artificial labs) and Predictive models (expert-done calculations that predict time to complete a task on a finished system) are very expensive / time consuming to use

HCI has generated several approaches that offer suggestions for redesign, but they focus on representation and require extensive, detailed modeling.  There is no ONE approach that addresses all types of activity that can lead to constructive suggestions for improving the system or device, that avoids details and allows for the identification of similar problems down the road.  In HCI, no one model is perfect- each one has its own limitations, which is why we should know about the cognitive dimensions framework.

Alan Blackwell and Thomas Green propose a set of cognitive dimensions framework that will allow researchers and designers to 'talk together' (DISCOURSE) about evaluation.  It is much lighter and easier to use than the above two main types of evaluation.

Cognitive Dimensions Framework

- not a method; it is a set of discussion tools; used by the designers themselves

- attempts to IMPROVE DISCUSSION through a SHARED VOCABULARY

- main process is to DISCUSS TRADEOFFS of design decisions among designers by using a SHARED VOCABULARY

- Cognitive Dimensions Framework suggests a handful of basic 'Notational Dimensions' in the shared vocabulary:  VISCOSITY, HIDDEN DEPENDENCIES , ABSTRACTION LEVEL, PREMATURE COMMITMENT

 

Notational Dimensions (the SHARED VOCABULARY)
- each dimension describes an aspect of an information structure that is reasonably general

• Viscosity:  resistance to change
• Visibility:  ability to view the components easily
• Premature commitment:  constraints on the order of doing things
• Hidden dependencies:  important links between entities that are not visible
• Role-expressiveness:  the purpose of an entity is readily inferred
• Error Proneness:  the notation invites mistakes and the system gives little protection
• Abstractions:  types and availability of abstraction mechanism
• Secondary notation:  extra information in means other than formal syntax
• Closeness of mapping:  closeness of representation to domain
• Consistency:  similar semantics are expressed in similar syntactic forms
• Diffuseness:  verbosity of language
• Hard Mental Operation:  high demand on cognitive resources
• Provisionality:  degree of commitment to actions or marks
• Progressive Evaluation:  work-to-date can be checked at any time

Evaluation using the CD Framework

There are two steps to evaluation using the cognitive dimensions framework:

1. Decide what generic activities a system is desired to support.  Each generic activity has its own requirements in terms of cognitive dimensions.
2. Scrutinize the system and determine how it lies on each dimension.  If the two profiles match, all is well.

Tradeoffs

What is nice about the CD framework is that it eliminates design maneuvers in which one dimension is 'traded-off' against another.  However, there are certain relationships:  such as a way to reduce viscosity is to introduce abstractions.  Abstractions might reduce viscosity and increase visibility...

Questionnaires

Because CD Framework is so general, it has been used to structure questionnaires to get user-feedback on certain aspects (notational dimensions) of a system.  Because the notational dimensions are very general, feedback can be very useful and may be unanticipated by the designer (a good thing).

(top)

Chapter 6:  Users' Mental Models
- stems from cognitive psychology
- why?  because understanding users' mental models can enrich our understanding of the use of cognitive artifacts

Cognitive psychologists think the UMM is at the heart of understanding HCI- reasoning behind what we do

However, the term is over used in so many different ways that it has lost its usefulness

Still, mental models is a very important and useful construct, with many areas that can still be researched

Cognitive psychology foundations to mental models:

Idea 1:  Mental content vs. cognitive architecture:  mental models as theories

• we use 'bounded rationality' - a process of using readily-available knowledge to make a decision.  (Note not ALL available information, because we make a quick decision and don't dig too deep for info)

  • cognitive psychologists focus on the 'structure' (human information-processing architecture) of the mind more than the 'content' (focus on what an individual knows / believes) of the mind to explain bounded rationality because 'structure' is more generalizable than content (individual / incidental).
     
  • while most cognitive psychology theories focus on structure of the mind rather than content of the mind (to achieve generalizability).  However, in HCI, we need to focus on singular behaviors (content) as they are a staple of usability studies, etc.  For example, a software 'walkthrough' will lead to diagnoses about misleading features of an interface design.
     

• users use mental models of familiar things to construct mental models of unfamiliar things (predicting the behavior of an ATM is based on other, familiar computer usage)

Idea 2:  Models vs. Methods:  mental models as problem spaces

• a characterization of skill:  a collection of methods for achieving tasks
   
• in cog. psych, solving a problem involves searching through a mentally constructed problem space of possible states (a.k.a. mental simulation)
   
• as we gain skill, we know what situation calls for what method (less searching / problem solving, more routine)

Idea 3:  Models vs. Descriptions:  Mental models as Homomorphisms

• mental models are 'analog representations' (share the structure of the world it represents)
   
• similar to pictures, 'simple static models' represent the world, with objects / relations
   
• 'dynamic models' are homomorphisms (one to many relationships rather than one to one), and can resemble state-transition networks

Idea 4:  Models of Representations:  Mental models can be derived from language, perception or imagination

• mental model can result from processing language, from a perception or from the imagination

Idea 5:  Mental representations of representational artifacts

• 'goal space' is the domain of the users' goals (such as spaces and lines on a paper when we are using a word processor)
   
• 'device space' is the problem space that the user searches for solutions to the goal (such as characters / operations in a word processor)

Idea 6:  Mental models as computationally equivalent to external representations

• it is possible to have two or more representations of the same information / task, but these are 'informationally equivalent' if one is inferable from the other and vice versa
   
• two or more 'informationally equivalent' representations may or may not be 'computationally equivalent', which is the cost of accessing the information / doing the task
   
• a 'cognitive map' is a mental representation that allows users to navigate through an environment (computationally equivalent to an external map) through process of internalization

Case studies:

What are 'yoked' state spaces? 

'yoked state spaces' can motivate an informal analysis of the fit between the representational capacities of a device and the purposes of a user (calendar / appointment diary example)

support the process of internalization - computationally equivalent mental versions of external representations

(top)

Chapter 7:  Exploring and Finding Information

How do people forage for information?  People are informavores (George Miller):  we are organisms hungry for information about the world and themselves.  This chapter draws much from evolutionary theory.

Two main concepts of this chapter:

• Information-Foraging Theory
• Information Scent

Information-Foraging Theory

• a metaphor from evolutionary theory:  deals with understanding how user strategies and technologies for information seeking, gathering, and consumption are adapted to the flux of information in the environment
• built on notation of hunter / gathering

Information Scent

• concerns the user's use of environmental cues in judging information sources and navigating through information spaces

Both ideas evolved in reaction to the technological developments associated with the growth of globally distributed, easily accessible information, and the theoretical developments associated with the growing influence of evolutionary theory in the behavioral and social sciences.

Following the metaphor:  What do we do when we hit a roadblock (metaphorically like hitting a stream, where we loose our 'information scent')?  We need to search around to pick up the scent again- much of the probability of success depends on the interface design (whether it's good or not- and can lead us in the right direction:  provide clues as to which choice to choose next, eliminate bad paths, etc.)  If we loose track of our 'scent' and accidentally pick up the wrong scent, we loose trust in the system.  Also, how far down the wrong path (scent) will we go before we realize we are on the wrong path?

Adaptation vs. Exaptation

this chapter's approach sees users as being adaptable.  Users are complex adaptive agents who shape their strategies and actions to be more efficient and functional with respect to their information ecology.

Adaptationist approaches became mainstream in the 1980s, as a reaction to ad-hoc models on human cognition (cognitive & perceptual tasks).  This is in contrast to mechanistic approaches of the time, such as the MHP (Model Human Processor, Card, et. al 1983)

Adaptationist approaches reverse-engineer the problem- examining what environmental problems are being solved and why cognitive and perceptual systems are well adapted to solving them

Exaptation looks at how we as humans are capable of adapting to solve similar problems (a way of generalizing our knowledge:  we can solve similar but different problems by observing themes in the problems)

Extending the Information Foraging Theory Metaphor:

• exaptation of food-foraging mechanisms and strategies of information foraging:  natural selection favored our ancestors- those with better foraging strategies had better success
• the economics of attention and the cost structure of information:  a wealth of information creates a poverty of attention- we need to efficiently allocate our attention to RELEVANT information.  Information systems should evolve to provide more valuable information per unit cost.
• relevance of optimal-foraging theory and models:  optimal foraging from study of animals- who try to get the most nutritious food with the least amount of effort.  With information, we want the richest, most relevant information in the least amount of effort

Optimal-Foraging Theory

Goal is to optimize the information we receive to be the most rich, relevant to our information needs.  Optimization models include three major components:

• Design assumptions:  specify the decision problem to be analyzed (how much time to spend processing a collection of information)
• Currency assumptions:  identify how choices are evaluated (information value = currency)
• Constraint assumptions:  limit / define the relationships among decision / currency variables (rise out of task structure / interface / knowledge of population)

In general, all tasks can be analyzed according to the value of the resource currency returned and costs incurred.  Classified as Resource costs (cost to get it) and Opportunity costs (potential benefit of it)

Scatter / Gather

• an interaction technique for browsing large collections of documents
• we cluster documents, then make a cluster hierarchy
• very messy looking interface on p. 170 using this- bad HCI!
• gather resources, scatter into categories

As Designers:  we want to design systems that provide the richest information, with the least cost to access it (easy information retrieval) - can be very useful in web design and search engines

Current Status of IFT:  Internet Ecology is being studied- looks at complex global phenomena that yield predictions on Internet usage and distribution of users over web sites

(top)

Chapter 8:  Distributed Cognition

Almost every work situation requires someone working with other people

HCI hasn't had a way into this problem because cognitive theory tells us little about social behavior, and social science is hard to apply directly to design

Ways in which people use tools (artifacts) to support their goals is poorly understood

Distributed Cognition grew out of a need to understand how information processing and problem solving could be understood as being performed across units LARGER than the individual
- cognitive scientists (psychology) don't have to abandon their background to understand this:  we just shift from focusing on 'information in the head' to 'information in the world' - examine things 'in the wild'
- how to understand how intelligence is manifested at the systems level rather than the individual level

Designing Collaborative Technologies

CSCW (Computer Supported Collaborative Work) shifts the focus of HCI from the user to the group (multiple, codependent users and their social network)

we focus on the problem solving of the group as a cognitive problem

we focus on distributed cognition within a context - drawing on actors and other features within the environment that allow problem solving (socially distributed cognition)

the goal of analysis:  describe how distributed units are coordinated by analyzing interactions among individuals, the representational media used, and the environment that the activity takes place

shift from traditional HCI (micro-structure level) to macro-structure level (ecological design - a turn to the social)

this has led ethnography - an anthropological approach to collecting data about the problem domain - to become a central feature of CSCW (see Ch. 13)

Cognition

• definition:  referring to all of the processes by which the sensory input is transformed, reduced, stored, recovered, and used
• cognitive science:  the problem solving and the organization of knowledge about the problem domain
• The dominant paradigm for cognition, Information-Processing Theory, proposes that problem spaces (the representation of the operations required for a given task) are abstract representations.  There is no reason this theory needs to be restricted to an individual - any unit performing these activities can be described as a cognitive entity
• artifact:  any man-made or modified objects (Cole, 1990)
• cognitive artifact:  a subset of artifacts - those that aid cognition ('knowledge in the world' - Norman; memory aids, etc.).  These can increase human potential by extending our abilities, and can also transform the task itself- which can allow the user to re-allocate resources for more efficient task completion
• with distributed cognition, this re-allocation of resources can take place across the group of individuals, for even more efficient task completion
• this has led to new theories on "systems perspective" that focus on describing the features of a system:  the people, artifacts, and the means of organizing these into a productive unit
• external cognition:  described by Preece as a way of externalizing information (creating and using information in the world, without performing logic operations in the head), in order to reduce memory load, to simplify cognitive effort by computational offloading onto external media, and by allowing us to trace changes through annotation or by using a new representational form.  This allows humans to logically process information without performing logic operations in their heads.
• cognitive ethnography:  analyzes the workplace to determine the role of technology and work practice in system behavior

Distributed Cognition and Computing

• looks to examine the role between the computer and the social group
• How is information about a domain stored (represented) in a system?
• functional system:  the actors and artifacts of the system (complex cognitive system / functional unit) - derived from the activity system (activity theory)
• make use of abstraction to describe a cognitive system - away from detailed design- focus on the general functions of the system (nonspecific:  what it DOES rather than what it IS)
• A distributed cognitive system consists of:  (see Fig. 8.2 - p. 207)  (note:  somewhat resembles MHP)
  • sensory mechanism - input from outside the cognitive system - passes to information processing unit
  • action generator - allows production of output from the cognitive system, can provide feedback to the information processor
  • memory - a stored representational state to order subsequent activities, receives representations from the information processor and passes back to it
  • information processor - receives representations from the sensory system and acts upon them:  transforms them, combines them or destroys them
• when a cognitive system is distributed over a number of individuals, cooperation is required among the individuals in order to bring their problem-solving resources into conjunction with each other.  This is crucial to their cooperative action
• a problem faced by groups in performing distributed cognition is the ability to organize a task into component parts that can be performed by individuals- and then re-assembled at the end (distribute the load fairly evenly across group members)
• we need to proactively structure distributed group labor so that effective work can be done
• there needs to be effective communication and coordination between group members (artifacts with universal meanings, etc)
• DCog vs. Ethnomethodology:  Ethnomethodology informed ethnography focuses on the ordering of the activity (coordination), but not on the work itself

Doing DCog:  Cognitive Ethnography

Unit of analysis:  the functional system (individuals, artifacts, and their relations)

Look for information-representation transitions that result in the coordination of activity and computations:

• examine the way that the work environment structures work practice,
• changes within the representational media,
• the interactions of the individuals with each other,
• the interactions of the individuals with system artifacts

(top)

Chapter 9:  Cognitive Work Analysis (sometimes called sociotechnical work analysis)

Cognitive engineering - the analysis, modeling, design and evaluation of complex sociotechnical systems
- first coined by Norman after Three-Mile Island crisis, helped launch the field (how to design human-machine systems so they are safer / more reliable)
- goal was to design systems with good user interfaces, so that in unanticipated situations, the user would be able to safely handle the physical system

Cognitive Work Analysis (CWA) - way of analyzing human cognitive work - describes the forces that shape human cognitive work in complex real-time systems (where humans control complex physical processes).  The goal is to lead to systems that better support human-operator adaptation when operators are confronted with unanticipated variability.  CWA is a multidisciplinary approach to cognitive engineering.

Often, the human operator is very separated from the actual process he is managing, due to a poor interface.  CWA is an approach to cognitive engineering that aims to help find better ways of connecting the operator with what he is managing (physical system behind the technology) - so in unexpected situations, adaptive behavior (by the operator) will be successful.

Ecological Interface Design (EID) - subset of CWA - a set of principles to guide interface design (influenced by ecological psychology).  Said to be one of the most useful products of CWA.  Makes use of WDA and WCA (see below), as well as some ecological concepts.

formative model - an approach that describes the requirements that must be satisfied so that a system can behave in a new, desired way.  CWA is a formative approach to work analysis.

Phases of Analysis in CWA:  Based on Figure 9.3, p.230
 - each narrows down the action possibilities further from the previous phase

1.  Work Domain Analysis (WDA):  find purpose and structure of work domain; often represented in abstraction-decomposition diagrams and abstraction hierarchies
 - constraints (physical & purposive) within which activity takes place (NOT the activity itself, though)
 - often 5 levels of abstraction:  functional purpose, abstract function, generalized function, physical function, and physical form (see Figure 9.7 on p. 242 for an example)

2.  Control Task Analysis (CTA):  find what needs to be done in the work domain so that it can be effectively controlled; often represented in maps of control task coordination, decision ladder templates
 - continues to narrow down the 'dynamic space of functional action possibilities' by defining constraints that must be satisfied when work functions are coordinated over time and when effective control is exercised over the work domain

3.  Strategies Analysis (SA):  find ways that control tasks can be carried out; often represented in information flow maps
 - 'HOW control tasks can be done' is analyzed (we don't care by whom)
 - focus on general classes of strategies and their intrinsic demands

4.  Social-Organizational Analysis (SOA):  find who carries out work and how it is structured; often represented by annotating the information flow maps of #3
 - focus on the division and coordination of work (the content (information passed between actors)), and the social organization of the workplace (the form (behavioral protocol of communications))

5.  Worker competencies analysis (WCA):  find the kinds of mental processing supported; often represented by skill-based, rule-based, and knowledge-based behavior models

CWA was influenced by systems thinking and ecological psychology.  Both emphasize that the human-environment system needs to be the unit of analysis, with the environment being a primary unit of analysis in an actor's goal-oriented behavior.

Systems Theory:
- the whole is more than the sum of its parts (study the whole environment)
- study the relation between parts, not the properties of the parts
- study cybernetics - open vs. closed systems - and how outside disturbances affect the system

Ecological Psychology:
- CWA wants to build 'ecological information systems' that can be operated closer to the ease which the natural world is navigated
- EID (ecological interface design) is an approach to building interfaces using principles of CWA
- "ecological science rests on the principle that systems in the natural and social world have evolved to exploit environmental regularities" - Rosson & Carroll
- ecological psychology is the study of the way organisms perceive and respond to regularities in information
- key concept #1:  environments and information should be described in terms that reveal their functional significance for an actor rather than being described in objective actor-free terms
- key concept #2:  the affordances of an object are the possibilities for action with that object, from the point of view of a particular actor
- key concept #3:  the actor uses direct perception, which proposes that certain information meaningful to an actor is automatically picked up from the visual array

The CWA System Life Cycle (SLC):  requirements definition, modeling and simulation, tender evaluation, operator training, system upgrade, and system decommissioning.  This area may see much development in the next decade.

At the end of the chapter, many case studies are presented.  They fall into 2 categories:  display design, and evaluations of human-system integration.

(top)

Chapter 10:  Common Ground in CMC:  Clark's Theory of Language Use

HCI has come to encompass technologies that mediate human-human communication (chat, etc)

Production + Comprehension = Communication

Clark's theory of common ground:  we use existing common ground to develop further common ground and hence to communicate effectively

• Different types of common ground:
  • a proposition is common ground if:  all the people conversing know the proposition, and they all know that everyone else knows the proposition.
  • communal common ground:  "we both speak English.  we are professionals.  we live in the same town."
  • personal common ground achieved before the conversation:  "our joint purpose is to sign off on the plan."
  • personal common ground developed during the conversation:  "the door on the bedroom that faces south has to be moved.  We can see the bedroom facing south on the plan.  The plan can go to the builder."
• Formalizes collaborative activity as "joint action"
  • language is a joint action involving two or more people
• Describes the process by which common ground is developed through joint action
  • face to face conversation uses common ground to reduce the effort required to communicate
  • face to face conversation develops common ground
  • face to face conversation involves more than just words (non-verbal comm, etc.)
  • face to face conversation is a joint action
  • face to face conversation is "basic" (basis for understanding language behavior)

Grounding:  process of making sure that another person sufficiently understands you.  If not- use grounding.  Often facial expressions (non-verbal behavior) or questioning serves to notify us if other person needs more information (grounding)

Summary:  Clark's theory of language use is applicable to CMC (computer mediated communication).  The usage / application of this theory to designing systems will be more evident as time goes on- and will likely be very useful to new systems supporting video conferencing, asynchronous and synchronous communication, etc.  However, a very useful set of guidelines based on this has yet to be developed.

(top)

Chapter 11:  Activity Theory

History / Foundation for Activity Theory:

A basic perspective on HCI that came from Scandinavian roots in the early 1980's

an 'Action Research Approach' that focused on active cooperation between researchers and 'those being researched' - researchers entered into an active commitment to improve the situation of those they researched

roots in social psychology, industrial sociology and critical psychology.  Influence by the introduction of the personal computer, moving away from the mainframe.

There were many problems with existing HCI and system design at the time.  Much of the system design being done at the time consisted of heavy task analysis, with consideration of a generic novice user working alone.  A new theoretical foundation was needed!

Foundations of Activity Theory:

• analysis and design for a particular work practice with concern for qualifications, work environment, division of work, and so on;
• analysis and design with focus on actual use and the complexity of multi-user activity.  In particular, the notion of the artifact as mediator of human activity is essential;
• focus on the development of expertise and of use in general;
• active user participation in design, and focus on use as part of design.

Activity theory shares with ecological psychology the attempt to move away from the separation between human cognition and human action, and an interest in actual material conditions of human acting.  Activity theory differs by adding a notion of motivation.  Activity theory values hierarchical analysis, task analysis, etc.  Activity theory takes place on all levels at the same time, not in sequence (see below).  Activity theory moves away from the generic user.

Unit of analysis in Activity Theory:  motivated activity.  This is mediated by socially produced artifacts (tools, language, representations, etc.)

Activity is mediated through a computer or other tool

Activity theory does not assume a boundary between internal representations and external representations, like cognitive science.  Activity theory has a basic feature of unifying consciousness and activity.

Development:  The most distinct feature of Activity Theory is development.  When compared to other materialist accounts in computer science, the focus is on development.

Nardi's Five Principles of Activity Theory:

• object-orientedness
• hierarchical structure of activity
• internalization and externalization
• mediation
• development

Mediation is folded in with each of the other four principles, resulting in four categories of concerns:

• means / ends
• environment
• learning / cognition / articulation
• development

Focus and Focus Shifts:

Activity theory starts with a perspective / point of view - which yields the objects to work with.  However, the focus shifts that indicate the dynamics of the situation are the main point of concern in the analysis.

(see Index page for more on Focus Shifts)

Summary:

Nardi suggests that activity theory is a powerful descriptive tool rather than a predictive theory.  Carroll somewhat disagrees- because in this chapter concrete techniques were presented that show how it can be used to focus on computer-mediated activity (a.k.a. HCI)

(top)

Chapter 12:  Applying Social Psychology Theory to Group Work

              Study _______________________
                   /                      /|
        Approach/                      / |
                 /                      /  |
         Build  /______________________/   |
                |                      |   |
     Technical  |                      |   |
Infrastructure  |                      |   |
               F|      Variations      |   |
  Architecture O|       in CSCW        |   |
               C|       research       |   |
   Application U|                      |   |
               S|                      |   |
          Task  |                      |   /
                |                      |  / 
        People  |                      | /  
                |______________________|/
                      Group Size
            Individual ... Group ... Society

CSCW (Computer Supported Collaborative Work) is a subfield of HCI that aims to build tools to help group work, learning and playing.  It examines how groups incorporate tools into their routines and the impact of CMC on group processes and outcomes.

Two main goals of CSCW:

• to support distributed groups (distributed groups are typically not as effective as collocated members)
• to make collocated groups more effective (CSCW systems are designed to ameliorate some of Steiner's process loss

Elements of an input-process-output model of groups:

 

 

 

 

Production Outcomes:  multidimensional

• groups do better than individuals because of aggregation (different people contribute to the groups with unique resources

• groups do better than individuals because of synergy (the increase in effectiveness that comes about from joint action and cooperation

• group maintenance and member support:  groups need to have the capability to work together in the future (group maintenance) and support the needs of its individual members (member support)

Inputs:  both inputs and processes that group members use will determine the success of the group

• personnel:  diversity can be a mixed blessing - more viewpoints, but more arguments

• task:  McGrath states that tasks can be:  generative (e.g. brainstorming); intellective (e.g. correct answers); problem-solving (e.g. open ended ?'s).  Groups in knowledge work tend to be as good as the second best person and follow a 'trust supported ' heuristic... a process of aggregation and synergy

• technologies:  groups will be more effective if they have more qualified personnel and appropriate technology (applies to collocated and distributed groups)

Interaction Process:  the way group members interact can directly influence group outcomes and mediate the impact of inputs on the group. 

• focus is on communication (takes time away from group production) and can be characterized according to volume, content, structure and interactive features

• the right volume, content, structure and interactive features depends on the particular task

• uncertainty is a key feature that determines whether a particular technology will be appropriate

• technologies restricting communication are less acceptable if tasks are more uncertain

Process Losses

• Steiner:  being in a group degrades performance from what the members could be producing individually

• can be caused by mis-coordination (production blocking, schedule conflicts, misaligned goals, etc.)

• can be caused by reductions in motivation (slackers, not present for goal setting, being around other slackers, social loafing)

• production blocking is when people cannot get work done because they are busy listening and participating with other group members

• social loafing is when people think their efforts are being pooled with the efforts of other group members.  People tend to work harder individually or when they realize their contributions to the group are unique, or if they like the group (it's attractive / valuable)

• social pressure (such as evaluation apprehension) can cause production losses.  A way to lighten this is to introduce anonymity, but this must be considered carefully (a set of trade-offs)

Anonymity:  a way to offset social pressures, but can cause social loafing (trade-off)

CSCW researchers turn to the social-science literature outside their own field, and often consult this- such as ethnographic research- than experimental social psychology

Carroll says that CSCW research has been underexploited- a lot due to mismatching goals and values between HCI/CSCW research and social psychology research

social psychology has provided a rich body of research and theory about principles of human behavior, which should be applied to the design of HCI applications- especially those supporting multiple individuals who are communicating or performing a collaborative task

(top)

Chapter 13:  Studies of Work in HCI

Jonathan Grudin argued in 1990 that HCI was passing from the 4th stage to 5th stage:  from "a dialog with the user" to a "focus on the work setting"

This leads into the topics of this chapter, which get into ethnography, situated action, and ethnomethodology.

Ethnomethodology

• predominated by the emergent HCI concerns with the workplace
• utilizes an ethnographic / fieldwork approach
• is concerned with the analysis of work and the workplace
• ethnomethodology & conversation analysis

Many factors precipitated the adoption of CSCW:

• Lucy Suchman's "Plans and Situated Action" - move from the individual user to a computer placed in a social context.  Attacked many cognitive scientists that failed to take into account the social and cultural world.  She undermined the idea that we 'plan' what we're going to do- replacing it with an idea that action takes place within sociocultural contingencies that cannot be covered by a plan (owes much to ethnomethodology)
• Suchman demonstrated that work could be studied as part of the process of designing systems for the workplace (draws from ethnomethodology within sociology and conversation analysis)
• the Scandinavian Participatory Design movement:  led to the involvement of the workers in the design process, and emphasized:  the flexibility of work activities, and that work is 'accomplished' rather than a 'mechanical process' (Activity Theory?)

Overview:  A Paradigmatic Case

• there are a number of ways in which ethnomethodologically grounded studies of work have been applied to HCI.  One way is to analyze the impact that a system has on the work that is done in the setting into which it is introduced
• the study of work analyzes the methods through which a domain of work is organized by those who are party to it from the inside, and makes clear that modeling a workflow using formal processes as a resource only partially grasps the work that a system is meant to support.
• studying work reveals a domain of work practices and methods that are crucial for the efficient running of an organization (characterized as "hidden" work of organizations)
• often, this 'hidden' work needs to be revealed for a successful analysis and design

Scientific Foundations

Ethnography

• theorize work
• empirical
• labor process theory (describe work as the reproduction of capital labor- not looking at the content of work.  serves as a structural orientation)
• an Interactionist approach:  attempts to develop an understanding of work from the inside (the actual workplace)
• Studies of work within HCI and CSCW tend to employ an ethnographic approach.  (Anderson:  ethnography in HCI has really stood for "fieldwork" grounded in ethnomethodology)

Situated Action

• Suchman:  human action takes place within contingencies.  Contingencies are not able to project in detail how any one conversation may unfold before participants engage in the course of social interaction.  Social actions are situated within arrangements and interaction.
• Social action should be analyzed within the situations in which it occurs in order to find how a course of action has been put together

Ethnomethodology

• primary concern is with social order
• social order is essential:  people display their orientation to the ordered properties of actions and interaction, and use the orderly properties when carrying out their social conduct
• members' phenomena:  order is to be found; order is a matter for ordinary members of the society
• members' accomplishment:  order is the product of consensus brought about by reciprocally shared normative values and rules, where Marxist sociology has showed that order is the product of constraint where the people are subject to the power if a coercive state exercised over them by it's institutions
• Ethnomethodology has explored the in situ production of social order through two broad domains of interest:  Conversation Analysis and Ethnomethodological Studies of work.  Both are work methodologies.

Conversation Analysis

Turn taking in organization is organized by participants on a moment-by-moment basis.  Turn taking is an interactional phenomenon, relating to multiple participants and organized as a collaborative matter.

Conversation analysis has developed as the study of turn taking in conversation.  It underpins some of the ways to study work, such as in the development of natural language interfaces.

Ethnomethodological Studies of Work

The second major preoccupation of ethnomethodology is its interest in work.  Ethnomethodological studies of work attempt to examine domains of work in order to understand what are the particular constitutive features of work, and how in their actions with one another, people are recognizably engaged in doing their work.

Critique:

• work is not organized as 'rule following' but rather in the contingencies and improvisations of applying rules
• Suchman critiqued a workflow system based on speech acts (Searle)- said that conversational analysis can underscore the situated and unfolding character of conversational exchanges

Summary:

In the past 10 years, there has been a shift in HCI from the user to the social world (work setting) in systems design.  The door has been opened to study the work setting, and information being gathered from the work place will allow us to design better systems.

(top)

Chapter 14:  Upside-Down Algorithms - Computational Formalisms & Theory

This chapter is about how to gain insight into the context of HCI design and evaluation by using theoretical concepts (computational theory) and methods (formal methods).  We must understand our raw material (the computer itself) and that drawings, sketches, etc. are part of the design process.

Formalism is about being able to represent things in such a way that the representation can be analyzed and manipulated without regard to the meaning.  An example of this is State Transition Networks.

Key Features of Formal Descriptions:

• formal analysis:  possible to ask questions about the system from the formal description
• early analysis:  analyze early by using rapid prototypes, etc
• lack of bias:  formal analysis helps to break biases
• alternative perspective:  different representations allow us to see different things about a design, providing more views of the artifact during design
• forcing design decisions:  using formal representation, the designer is forced to make user-interface decisions explicitly and communicate those decisions to the implementer

Detailed Description:

• Fitt's Law shows that no matter what the size or number of the screen buttons - that a reasonable typing speed is always faster (3x faster than mouse clicking)
• Touch screens:  large targets are better
• Per screen cost:  more items per screen the better... ???
• for small targets, small numbers of well explained items may be better

Reasons for using Formal methods in HCI:

1. to analyze the system to assess potential usability measures or problems
2. to describe the system in sufficient detail so that the implemented system is what the designer intends
3. the process of specification forces the designer to think about the system clearly and to consider issues that would be missed

Formal Modeling can be done for single user systems or for cooperative work

Case study:  flowcharts of the human-computer dialogue:  work well because they are simple formal methods.  Why:

• useful:  addresses a real problem
• appropriate:  no more detail than needed
• communication:  mini pics and clear flow are easy to walk through with client
• complementary:  different paradigm than implementation
• fast pay-back:  quicker to produce application
• responsive:  rapid turnaround of changes
• reliable:  clear code is less prone to errors
• quality:  easy to establish test cycle
• maintenance:  easy to relate bug / enhancement reports to specification and code

Summary:

There once was a time when every computer method had to be formal to be respectable.  However, formal models bred problems over time, and grew to the point where they were written off.  However, we should not write off formal methods - look at how successful UML diagramming has been.  Often using a formal model can be beneficial- especially those that projects simple in nature.

The web is an example of a rapidly growing distributed system.  Things like bookmarks and the 'back' button are stupid- it restarts the application in the middle of a process!

The problem and challenge of formal methods:  whenever we capture the complexity of the real world in formal structures, whether language, social structures, or computer systems, we are creating discrete tokens for continuous and fluid phenomena.  In doing so, we are bound to have difficulty (it is impossible to capture everything, in minute detail to a program, process, etc.).  However, it is only in doing these things that we can come to understand, to have valid discourse, and to design.  (think of Formal Models as 'naive physics' - as simple models that can work quickly / easily, but do not accurately represent reality)

(top)

Chapter 15:  Design Rationale as Theory

This chapter aims to show how reflective HCI design practices (involving design-rationale documentation and analysis) can be used to: 

1. closely couple theoretical concepts and methods with the designed artifacts that instantiate them
2. to more closely integrate theory application and theory development in design work
3. to more broadly integrate the insights of different technical theories

Design rationale contributes to theory development in HCI in three ways:

1. it provides a foundation for ecological science in HCI by describing the decisions and implicit causal relationships embodied in HCI artifacts
2. it provides a foundation for action science in HCI by integrating activities directed at description and understanding with those directed at design and development
3. it provides a framework for a synthetic science of HCI in which the insights and predictions of diverse technical theories can be integrated

Design Rationale

• is documentation and analysis of specific designs in use
• describes the features of a design, the intended and possible use of those features, and the potential consequences of the use for people and their tasks
• this involves observing or hypothesizing scenarios of user interaction, and describing their underlying design tradeoffs
• helps to make theory more applicable by codifying the terms and relations of the application domain, and grounding them in design tradeoffs and decisions

Applying theory in HCI design involves mapping concepts across domain boundaries, and directing descriptions and analysis to prescriptions for intervention.

TAF:  Task-Artifact Framework

• design rationale is created to guide and understand the impacts of computing technologies on human behavior
• Tasks analysis is expressed as scenarios of use.  The scenarios include the activity context, the actors' motivations, actions and reactions during the episode
• Claims analysis produces a causal analysis of the actors' experience, enumerating the features of a system-in-use that are hypothesized to have upsides and downsides for the actors.
• The design rationale for a system is built up through analysis of multiple scenarios, often over incremental versions, leading to a network of overlapping claims.
• Book exemplifies these through MOOsburg example - p. 434

Design Rationale:  Three scientific foundations:

1. Ecological science:  rests on the principle that systems in the natural and social world have evolved to exploit environmental regularities. 
   • HCI can be developed as an ecological science at three levels:
     1.  taxonomic science
     2.  design science
     3.  evolutionary science
2. Action science:  a principle to research that closely couples the development of knowledge and the application of that knowledge.  Integrates the traditional scientific objectives of analysis and explanation with the engineering objective of melioration.
3. Synthetic science:  the design rationale that surfaced during a design project can be grounded in existing scientific theory, or can instantiate predictions that would extend existing theory

Detailed Description:

• the most important step in constructing a design rationale is to identify a collection of typical and critical scenarios of user interaction
• Methods for identifying scenarios:
  • carry out the work that a system is intended to support
  • carry out the work that collaborative workshops with users envisioning alternative ways
  • draw on prior work or case-study reports of other design work
  • instantiate general types of interaction scenarios in the current domain
  • transform existing scenarios
• Methods to recognize claims (tradeoffs) in scenarios
  • Text analysis:  raise questions about the actions, events, goals, and experiences depicted in the scenario, and jointly analyze scenarios with users

Case Study:  MOOsburg

The design rationale associated with an interactive system can be evaluated and refined.  In the case of MOOsburg, prototypes are still being developed, analyzed, used and refined.

When a design rationale is generalized, the hypothesized causal regularities contribute to theory building.  Design rationale supports ecological science at three levels:

• supports taxonomic science by surveying and documenting causal regularities in the usage situation
• supports design-based science through abstractions that enable knowledge accumulation and application
• supports the development of evolutionary science by promoting insights and development of new features and new situations

When the generalized rationale is grounded in established scientific theory, it serves as an integrative frame within which to understand and further investigate competing or complementary concerns (synthetic science)

In general, scenarios and design rationale specify a shareable design space that can be used to raise, discuss, and arbitrate widely varying theoretical prediction and concerns.

 

(top)

Jump to Chapter:  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15