Methods in Perception Methods in Perception Research Chapter 1: Fundamentals of Sensory Perception • Qualitative: Getting the big picture. • Quantitative: Understanding the details • Threshold-seeking methods • Magnitude estimation others, often similar to those used in • Many other areas of psych. 1 2 Qualitative Observation Qualitative Methods in Perception • • “Thatcher Illusion” • Much quantitative work now trying to find out why it happens & what it means Qualitative observation uncovered the phenomenon Schwaninger, A., Carbon, C.C., & Leder, H. (2003). Expert face processing: Specialization and constraints. In G. Schwarzer & H. Leder (Eds.), Development of Face Processing, pp. 81-97. Göttingen: Hogrefe. 3 • Also called phenomenological or naturalistic observation methods • Relatively non-systematic observation of a given perceptual phenomenon or environment (e.g., an illusion) • Yields a verbal description of one’s observations, (possibly with some simple numerical assessment) • First step in any study of any perceptual phenomenon. Gives the “big picture” 4 Qualitative Methods in Perception • Example: Famous perception researcher Jan Purkinje noticed that his flower bed looked light red/dark green during the day but dark red/light green at twilight • This phenomenological observation led to the hypothesis of 2 visual systems, & ultimately to an understanding of the functions of rods and cones 5 6 Quantitative Methods Quantitative Methods in Perception seeking methods measure a • Threshold physical quantity representing a limit of perceptual ability (i.e., a threshold) in physical units (meters, decibels, • Measured parts-per-million, candelas of light, etc.) 7 • Absolute threshold - smallest detectable physical quantity (e.g., 2 dB, 3.57 grams...) • Difference threshold - smallest detectable difference between two physical quantities 8 Quantitative Methods in Perception are defined for a given level of • Thresholds response accuracy • Typically we speak of the “50% threshold”, meaning the physical quantity detectable (absolute threshold) or the physical difference detectable (difference threshold) 50% of the time a threshold can be defined for any level • But of accuracy (e.g., 75%, 83.6%...) Threshold-seeking Methods • Classic methods (Fechner, 1850’s) • Method of adjustment: Quick and dirty • Method of limits: Easy on observer, fairly fast and accurate • Method of constant stimuli:Very slow but very accurate • Adaptive methods: Fast, very accurate, but can be difficult for untrained observers 9 10 Method of Adjustment Method of Adjustment • Stimulus intensity is adjusted (usually by the Instructions: Adjust the intensity of the light using the slider until you can just barely see it observer) continuously until observer says he can just detect it • Threshold is point to which observer adjusts the intensity • Repeated trials averaged for threshold but not always accurate, due to • Fast, inherently subjective nature of adjustment 11 Photometer Reading: 0.9 0.1 cd/m2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 12 Method of Limits Method of Limits of different intensities presented in • Stimuli ascending and descending order (descending sequence) Instructions: For each light intensity, indicate whether you can detect it. • Observer responds to whether she perceived the stimulus • Cross-over point (between “yes, I see it” and “no, I don’t”) is the threshold for a sequence • Average of cross-over points from several ascending and descending sequences is taken to obtain final threshold. Photometer Reading: 0.9 0.2 cd/m2 0.3 0.4 0.5 0.6 0.7 0.8 13 14 Method of Limits (ascending sequence) Instructions: For each light intensity, indicate whether you can detect it. Example Data From Method of Limits • Why ascending and descending sequences? • Why different starting points? Photometer Reading: 0.9 0.2 cd/m2 0.3 0.4 0.5 0.6 0.7 0.8 15 16 Method of Constant Stimuli • 5 to 9 stimuli of different intensities are presented many times each, in random order Method of Constant Stimuli Photometer Readings Instructions: For each light intensity, indicate whether you can detect it. • The intensities must span the threshold, so must know approx. where it is a priori. trials (often 100’s) of each intensity • Multiple are presented is the intensity that results in • Threshold detection in 50% of trials 17 18 Question Here are some data from a participant in the MoCS. Can you estimate her 50% threshold? Stimulus Intensity cd/m2 % of stims detected 0.2 0 0.3 0 0.4 0.1 0.5 0.4 0.6 0.7 0.7 0.8 0.8 0.9 0.9 0.9 19 0.7 0.5 0.2 0.9 0.4 cd/m2 The Psychometric Function .2 .3 .4 .5 .6 .7 .8 .9 (Stimulus Intensity cd/m2) 20 0.4 0.7 0.4 0.9 0.4 0.2 0.5 0.5 0.5 . . . • To calculate thresholds we use curve-fitting techniques to fit a sigmoidal (=s-shaped) function to the data (green dots). • This is called a psychometric function (grey line). It links physical stimulus intensity to performance • Calculating 50% Threshold Intensity Math Review: Logs • Remember exponents? 2 = 4 • Logarithms are the opposite log (4) = 2 exponent will turn the base--usually 2, • “What 10 or e--into the operand?” 2 2 Debate exists over which kind of function--Cumulative Normal, Weibull, Logistic, etc.-is theoretically best, but in practice differences are minor • What is the log (100)? log (8)? • Khan Academy: http://tinyurl.com/7zy5wu7 10 2 21 22 Math Review: e (or “Euler’s Number”) Math Review: Inverse Functions • A mysterious constant that just pops up everywhere in nature. e ≈ 2.71828... • log is called the “natural logarithm”, often symbolized as ln e • One also sees ex, where x can be quite a complex expression. “exp(x)” is also used. to use your calculator to do logarithms • Learn and work with e 23 • If a function F takes input X and returns Y then inverse F takes input Y and returns X • Example:Y = 2X inverts to X = ½Y y = 1.0 - exp(-(x/b) ) • The Weibull: • The inverse Weibull x = b(-ln(1-y)) a (1/a) 24 Math Review: Free Parameters Math Review: Free Parameters y = 10 +1x • Free parameter: Part of a math function that is adjusted so that the function fits data • Example: The intercept (a) and slope (b) of a line are free parameters when fitting a regression line: y = a + bx; y = 0 +2x y = 0 +1x y = 1.0 - exp(-(x/5.0)1.0) • In the Weibull (and its inverse) a and b are free parameters. • a is the “offset” (in this example a = 1 in all cases) • b is the “slope” (here varying from 0.5 to 5.0) y = 1.0 - exp(-(x/1.5)1.0) y = 1.0 - exp(-(x/1.0)1.0) y = 1.0 - exp(-(x/0.5)1.0) 25 The Weibull Function y = 1.0 - exp(-(x/b)a) • • • • x is stimulus intensity (in some positive physical unit) y is predicted probability of stimulus detection (from 0 to 1) a is the “offset” and b is the “slope” a and b are free parameters whose values are chosen so that the curve best fits the data. These values are determined by curve-fitting algorithms whose details are beyond the scope of the course. 27 26 Calculating 50% Threshold Intensity Via Inverse Weibull x = b(-ln(1-y))(1/a) • To figure out the threshold, we need to figure out what the right values of a and b are for the Weibull that best fits our data. will then enter the percentage threshold • We we are seeking (e.g., .5 for 50% threshold) into the inverse Weibull (above) to determine the associated stimulus intensity x 28 Inverse Weibull step-by-step Calculating 50% Threshold Intensity Open the excel file "WeibullFit.xlsx" • Click the Tools menu and select "Solver…". Enter your data in two columns: X (stimulus intensity) in column A Y (proportion detected) in column B x = .603 (-ln (1-0.5) )(1/4.08) Note that you may have to first activate the solver. See the document on the class website called “Loading MS Solver Add”. • In the solver window, click “Options”, then enter “10” in “max time”, then click “Okay”. • Click "Solve" in the window that appears. After a moment, the Fit Parameters will appear in cells G2 (slope, or A) and G3 (offset, or B) • 0.9" 0.8" 0.7" x = .603 (.6931)(.245) x = .603 (.9141) 0.6" 0.5" "Data" 0.4" "Fit" 0.3" 0.2" 0.1" So for our data, the inverse Weibull is x = .603 (-ln (1-y) )(1/4.08) x = .5512 0" 0.00" 0.20" 0.40" 0.60" 0.80" 1.00" X" Plug in the desired probability of 0.5 and to a threshold of 0.5512 cd/m2 29 30 Inverse Weibull step-by-step Example for Self-Test Here are some data from a participant in the MoCS. What is her 50% threshold? What about her 83% threshold? Stimulus Intensity x = .603 (-ln (1-0.5) )(1/4.08) 1" x = .603 (.6931)(1/4.08) 0.9" 0.8" Y"(Data)" 0.7" x = .603 (.6931)(.245) x = .603 (.9141) 0.6" 0.5" "Data" 0.4" "Fit" 0.3" 0.2" 0.1" x = .5512 0" 0.00" 0.20" 0.40" 0.60" X" 31 0.80" 1.00" % of stims detected 0 0 1 0.05 2 0.15 3 0.45 4 0.75 5 0.85 6 0.95 7 0.95 32 50% =c.bj 83% = d.fc • 1" x = .603 (.6931)(1/4.08) Y"(Data)" • • Adaptive Methods Questions • Examples: Staircase, QUEST, PEST, etc. • Stimulus first presented at an arbitrary level • If observer perceives it, intensity is reduced by a predetermined step-size • What is a “free parameter”? • What is log (10000)? • If observer does not perceive stimulus, intensity is increased 10 • This is repeated until several reversals are obtained. • Threshold is average of reversal points. 33 34 Stimulus Intensity Reversal Points 700 600 700 500 600 400 500 300 400 200 300 100 0 200 N N N N N N Y Y Y N N N Y Y Y N N Y Y N Observer Response “Do you see it?” Threshold = (700 + 400 + 700 + 400 + 600 + 400) / 6 = 533.33 35 100 0 N N N N N N Y Y Y N N N Y Y Y N N Y But why is it called the “Staircase Method”? 36 N N Stimulus Intensity Example for Self-test Adaptive Methods 700 600 500 • Similar to method of limits, but we don’t run the same series of levels each time, instead “adapting” each series based on previous performance. • A number of specific procedures exist, which modify how the adaptation is done (size of steps, etc.): Staircase, QUEST, PEST, etc. • Very efficient, because almost all trials are close to the threshold and therefore informative. • But best used with trained psychophysical observers (can be frustrating for untrained Ss) 400 300 200 100 0 N N N N N Y Y Y N N N N Y Y Y N N Y N Y Observer Response “Do you see it?” Calculate the threshold for this run by this observer. ebh.f 37 38 Forced Choice Variations X-Alternative Forced Choice Variations • Some threshold-seeking methods can be hampered by response bias. • Some participants have a lax criterion and tend to say “yes, I see it” a lot. participants have a strict criterion and • Some tend to say “no, I don’t see it” a lot. way to mitigate this problem is to use • One forced-choice methods. 39 a forced-choice task, a participant is given • Inseveral options to choose from: One contains the stimulus and the others don’t. • The participant is asked, for example: “Which of the two boxes has a light in it?” • Note that a forced choice method is a modification of (addition to) the thresholdseeking methods we’ve looked at so far. 40 X-Alternative Forced Choice Variations number of alternatives can be offered, • Any but usually 2 to 8 are given. 2AFC Method of Limits (descending sequence) Instructions: For each light intensity, indicate which side it’s on. • A “2AFC” is a “two-alternative forced choice” procedure, for example. • The number of alternatives affects the “chance level performance” (= 100% / A) Photometer Reading: 0.9 0.2 cd/m2 0.3 0.4 0.5 0.6 0.7 0.8 41 42 Threshold-finding With Other Senses Questions same methods can be used with a wide • The variety of sensory qualities. • Different physical units are used depending on the modality: • Sound amplitude (decibels) • Touch pressure (pascals) • Smell/Taste intensity (parts-per-billion) 43 • What is an absolute threshold? • Name several methods for measuring absolute thresholds. • What is an xAFC procedure? Why use one? generally how the method of limits • Describe works. 44 Difference Threshold • Smallest intensity difference between two stimuli a person can detect • Produces a “Just Noticeable Difference” Weber’s Law • (JND) in degree of subjective sensation • Same psychophysical methods are used as for absolute threshold • As magnitude of stimulus increases, so does difference threshold (∆I) 45 Weber’s Law describes the relationship between stimulus intensity (I) and difference threshold (∆I) as follows ∆I / I = K or ∆I = I × K • This holds true for many senses and many physical quantities across a wide range of moderate intensity levels (but see Steven’s Power Law) 46 Weber’s Law 47 48 2AFC Method of Limits for ∆I Method of Adjustment for Difference Threshold Instructions: Adjust the intensity of the test light using the slider until you can just barely see the difference between it and the standard light Standard Light: Photometer Readings 0.6 cd/m2 Test Light: 0.1 cd/m2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Instructions: For each set of lights, indicate which pair (right or left) differ. 49 Standard Light (Constant) 50 Test Light (Adjustable) 20 • It Q: If your Weber fraction (k) is .05, and the standard light’s intensity (Is) is 20, what level will the test light intensity (It)have to be raised to in order for it to be just noticeably different from the standard? Standard Light (Constant) Test Light (Adjustable) 50 It A: It will have to be higher by the ∆I; That is, It = Is + ∆I where ∆I = Is × k ∴ It = Is + (Is × k) It = 20 + (20 × .05) = 21 51 • Q: If your Weber fraction (k) is .20, and the standard light’s intensity (Is) is 50, what level will the test light intensity (It)have to be raised to in order for it to be just noticeably different from the standard? 52 fj.j • Example for Self-Test Questions Beyond the Threshold… • A participant can just tell the difference between lights of 100 cd/m2 and 112 cd/m2. Therefore, she should be able to just tell the difference between 200 cd/m2 and ____ cd/m2. • What is this participant’s Weber fraction for light intensity? 53 Beyond Thresholds Fechner’s Law • What do we know about the relationship • • Classic work done by Fechner, who derived From Weber’s findings, Fechner derived the idea that subjective sensation (S) related to stimulus intensity according to: S = k × ln(I/I0) • • • • k = empirically-determined free parameter between subjective sensation and objective intensity at levels above threshold? Fechner’s Law from Weber’s work. has since been largely supplanted by • This Steven’s Power Law. 55 Recall that “ln” means “log to base e” I = stimulus intensity I0 = stimulus intensity at absolute threshold 56 Fechner’s Assumption Fechner’s Assumption assumed that each time you went • Fechner up by one difference threshold (or, subjectively, one JND), that that related to a an equal jump in subjective intensity. • Intuitively appealing, but turns out to be wrong, as Stevens later showed. 58 Fechner’s Assumption Fechner’s Law Subjective Sensation (S) 57 • Works fairly well for some modalities (loudness, weight), but not others (electric shock) that show response expansion • More closely models the response of individual neurones than people. ∆I ∆I ∆I your music player’s volume control is • Still, modelled after Weber’s and Fecher’s laws Objective Stimulus Intensity (I) 59 60 Magnitude Estimation • • • • Technique pioneered by Stevens to examine the relationship between subjective perception and objective stimulus intensity at easily perceived levels • • All stimuli are well above threshold • Observer assigns numbers to the test stimuli relative to the standard (e.g., “that looks about twice as bright, I’ll call it a ‘10’. ”); • Relationship between intensity and perception usually shows either: • Response compression: As intensity increases, the perceived magnitude increases more slowly than the intensity Observer is given a standard stimulus and a value for its intensity (e.g., “see this light? this is a ‘5’.” ) expansion: As intensity increases, • Response the perceived magnitude increases more quickly than the intensity 61 62 Magnitude Estimation Magnitude Estimation Relationship between intensity and perceived magnitude is a power function Stevens’ Power Law • Magnitude Estimation S = k × Ib (amazingly simple!) Note how for b < 1, we get an approximation of Fechner’s law (red line) 63 • • • • S = k × Ib • Note that k is not the Weber Fraction S = perceived magnitude I = physical intensity k and b are empiricallydetermined free parameters 64 The 3 Laws of Psychophysics • Weber’s law relates two physical units: Standard stimulus intensity and difference threshold • Fechner’s law relates a physical unit (stimulus intensity) with subjective sensation (or so he thought). • • The above two derive from one another directly, and are sometimes called the Weber-Fechner law Steven’s Power Law expands on Fechner’s Law, covering stimuli that show response expansion as well as more closely modelling human responses. 65 Sensitivity & Signal Detection Theory 67 Questions Steven’s Power Law: S = k × Ib Given that for bitterness b=2 and k=2, what will sensory magnitude (S) be for stimulus intensity (I) of 2 ppm? What if I is doubled to 4 ppm? What does this mean about the relationship between stimulus intensity and sensory magnitude in this case? 66 Sensitivity & Signal Detection Theory • Sensitivity (d’) • Criterion (c) spend some time on this, as it is used in • Will many fields (though with different jargon): • Diagnostics • Inferential statistics • Human factors engineering 68 Absolute Thresholds Ain’t So Absolute • Thresholds shift for many reasons unrelated to actual sensitivity. • An especially problematic factor is criterion (tendency to say “yes” a lot or “no” a lot). • For example, an individual doing Method of Limits could just say “yes” to all stimuli (lax criterion) and look like he’s incredibly sensitive! • xAFC methods are one way to deal with this, but another is to measure Sensitivity instead of threshold. Sensitivity • Sensitivity is symbolized d' (“dee-prime”). • Measure of one’s ability to detect a given signal (= a stimulus), usually at low intensity. • Arises from Signal Detection Theory (more later) • How might we measure sensitivity? 69 70 (a bad way of) (a good way of) Measuring Sensitivity Measuring Sensitivity • Present stimulus 100 times and note % of times participant says he detects it? P who says “yes” all the time will • Problem: do very well. (the very problem we are trying to avoid!) • Need measure that reflects ability to discriminate between “signal present” and “signal absent” 71 • Present stimulus (signal) on only half of the trials, (test trials). • Present no stimulus (noise) on other half of trials, (catch trials). • Note that unlike threshold experiments, we present only one stimulus at one intensity. • Many trials are presented. For each, the participant says “yes, the stimulus is there” or “no, it isn’t”. 72 Four Possible Results on Each Trial Also Known As... Present Absent Yes, I see it Hit False Alarm No, I don’t Miss Correct Rejection The difference is really... Statistical test says... Participant says... The stimulus is really... Yes, it’s there No, it’s not Present Absent True Positive False Positive (type I error) False Negative (type II error) True Negative 73 74 Some Example Results No, I don’t Absent (n=100) 90 20 10 Benny Participant says... Yes, I see it Present (n =100) 80 Present (n =100) Absent (n=100) Yes, I see it 80 10 No, I don’t 20 The stimulus is really... Claire Participant says... Participant says... Albert Sensitivity The stimulus is really... The stimulus is really... Present (n =100) Absent (n=100) Yes, I see it 90 30 No, I don’t 10 70 75 results of such an experiment yield: • The proportion of hits (P = N / N ) h 90 hits testtrials proportion of FAs (Pfa= Nfa / Ncatchtrials) example, for Albert: • For P = 90 / 100 = .9 h Pfa= 20 / 100 = .2 • Note that we could calc proportions of misses and correct rejections too, but these are redundant (Pm = 1-Ph; Pcr = 1-Pfa) 76 Questions Sensitivity • Perfect participant: P = 1, P = 0 • Participant just guessing: P = .5, P = .5 • Worst possible participant (perfectly h • What is Benny’s proportion of hits (P )? • What is his proportion of FA’s (P )? • How do sensitivity experiments differ in h fa method from threshold-seeking methods? h fa backwards) : Ph = 0, Pfa= 1 • In calculating sensitivity, we want to reward hits and punish FAs, so we could just use “Basic Sensitivity”: BS = Ph - Pfa 77 78 we calculate • Instead d' = z(P ) - z(P ) Sensitivity • • • • • fa h FA Careful! This means function z, do NOT multiply z by PFA) the proportions of hits and FAs • Converting to z-scores yields a more valid result. BS equals: Perfect participant: Ph of 1 - PFA of 0 =1 Participant guessing: Ph of .5 - PFA of .5 = 0 Backward participant: Ph of 0 - PFA of 1 = -1 So BS seems to work. However, for obscure statistical reasons, BS is, well, B.S. 79 • d’ is measured in standard deviation units. • How to calculate the z scores? • In Excel, use norminv(P, 0, 1) • Table A5.1 from MacMillan & Creelman • Or the “unit normal” table from any stats textbook. 80 Normal distribution for finding d' (based on Macmillan & Creelman (2005), Signal Detection: A User’s Guide) Sensitivity • • d' = z(Ph) - z(Pfa) • Ben: Ph = .8 ∴ z = 0.84; Pfa = .1 ∴ z = -1.28 ∴ d’ = 0.84- (-1.28) = 2.12 • Claire: Ph = .9 ∴ z = 1.28; Pfa = .3 ∴ z = -0.52 ∴ d’ = 1.28 - (-0.52) = 1.8 Albert: Ph = .9 ∴ z = 1.28; Pfa = .2 ∴ z = -0.84 ∴ d’ = 1.28 - (-0.84) = 2.12 81 (MacMillan & Creelman, 2005, chp. 1). For our purposes, just substitute values of 0.01 and 0.99, respectively. 83 The stimulus is really... Present (n =100) Absent (n=100) Yes, I see it 0 0 No, I don’t 100 100 Flawless Fran Participant says... Doubting Dan Participant says... • What to do with proportions of 0 and 1? • These yield z scores of –∞ and +∞ • There are many ways of getting around this 82 Participant says... Zero and One Link back to data (#91) Easy Eddie Present (n =100) Absent (n=100) Yes, I see it 100 100 No, I don’t 0 0 The stimulus is really... Present (n =100) Absent (n=100) Yes, I see it 100 0 No, I don’t 0 100 84 The stimulus is really... Sensitivity • Eddie: Ph = 1 ∴ z = 2.33; PFA = 1 ∴ z = 2.33 ∴ d’ = 2.33 - 2.33 = 0 (usual minimum) • Dan: Ph = 0 ∴ z = -2.33; PFA = 0 ∴ z = -2.33 ∴ d’ = -2.33 - (-2.33) = 0 (usual minimum) Fran: Ph = 1 ∴ z = 2.33; PFA = 0 ∴ z = -2.33 ∴ d’ = 2.33 - (-2.33) = 4.66 (conventional max.) for self-test: Zack does a • Example sensitivity experiment with 20 test trials and 20 catch trials. He gets 10 hits and 5 false alarms. • What are his P • What is is d’? h 85 and Pfa values? d’ = .fg d' = z(Ph) - z(PFA) Ph = .ej Pfa= .be • • Questions 86 Criterion Criterion Link back to data (#91) • The flip side of sensitivity is criterion or response bias. • How do we measure a person’s tendency to say yes or no? There are several measures, but the simplest is: c = [z(Ph) + z(Pfa)] / -2 • The lower c, the more a person’s tendency to say “yes”. When c<0, yes > no. When c>0, yes < no. 87 • • c = [z(Ph) + z(PFA)] / -2 • Ben: Ph = .8 ∴ z = 0.84; PFA = .1 ∴ z = -1.28 ∴ c = (0.84 -1.28) / -2 = .22 [tends to “no”] • Claire: Ph = .9 ∴ z = 1.28; PFA = .3 ∴ z = -0.52 ∴ c = (1.28 - 0.52) / -2 = -.38 [tends to “yes”] Albert: Ph = .9 ∴ z = 1.28; PFA = .2 ∴ z = -0.84 ∴ c = (1.28 - 0.84) / -2 = -.22 [tends to “yes”] 88 Isosensitivity Curves (a.k.a. “ROC curves”) A range of Ph and Pfa values can produce the same d’. Pfa z(Ph) z(Pfa) 0.8 0.4 0.84 -0.25 1.09 -0.295 0.6 0.2 d’ Plotting results for different criterion levels at the same sensitivity yields an isosensitivity curve (aka, ROC curve) c 0.25 -0.84 1.09 0.295 1 ROC Curve d' = 1.09 0.8 d’ = 1.09, criterion = -0.30 0.6 Hit Rate Ph Isosensitivity Curves d’ = 1.09, criterion = 0.30 0.4 0.35 0.07 -0.39 -1.48 1.09 0.935 This occurs when sensitivity remains the same, but criterion shifts d’ = 1.09, criterion = 0.94 0.2 0 0 0 Hit Rate (Ph) 0 .5 0 0.5 False Alarm Rate (Pfa) 91 0.8 1 Why Does Criterion Shift? 0.0 1.0 1.0 Miss Rate (Pm) .5 0.6 90 Correct Rejection Rate (Pcr) 0.5 0.4 False Alarm Rate 89 1.0 1.0 0.2 • Many reasons, but an important one is the payoff matrix. • If the cost of a false alarm, or reward for a correct rejection, is great, one will tend to make one’s criterion stricter (more “no”) • If the cost of a miss, or reward for a hit, is raised, one will tend to make one’s criterion laxer (more “yes”) 92 Example of a Pay-off Matrix Leading to a Lax Criterion The stimulus is really... Participant says... Neutral Cancer is... Yes, I see it Patient Saved Unnecessary Additional Testing No, I don’t Patient Dies Yes, I see it +10$ -$10 No, I don’t -$10 +10$ The stimulus is really... The stimulus is really... Lax Patient Goes Home Absent (n=100) Present (n =100) Absent (n=100) Yes, I see it +100$ -$10 No, I don’t -$10 +10$ Strict Participant says... Absent (n=100) Participant says... Radiologist Says... Present (n =100) Present (n =100) Present (n =100) Absent (n=100) Yes, I see it +10$ -$10 No, I don’t -$10 +100$ 93 94 Questions Questions Dr. X tests a new field diagnostic procedure by applying it to 50 individuals known to have an illness. He finds that it correctly labels all 50 of them as ill, whereas his old procedure labeled only 40 of them as such. He concludes that the new procedure is better. Is his conclusion sound? Why or why not? 95 would one be interested in measuring • Why changes in criterion? • True or False: As one travels up and to the right along an isosensitivity curve, criterion rises. • True or False: A higher criterion value indicates a stronger tendency to say “yes, the stimulus is present” 96 Signal Detection Theory • The concepts of sensitivity and criterion arise out of Signal Detection Theory (SDT) • SDT suggests that any attempt at detecting a signal (stimulus) has to contend with competing noise • Noise in this sense is random variations from the environment (e.g., literal noise) or from within the detector (e.g., neuron chatter) SDT • • • Example:You’re in the shower and expecting a call. • If the phone isn’t ringing, you have just noise. If the phone is ringing, you have signal+noise. • d’ is essentially a measure of how similar the signal is to the noise for you subjectively. Signal: Phone ringing Noise: Sound of shower (plus all other sources of sound), as well as your own internal neuron chatter. 97 98 Probability Distributions of Noise and Signal+Noise Probability That a Given Perceptual Effect is Due to N or S+N noise only not at all phone + noise Probability phone + noise Probability noise only a bit kinda a lot unmistakably How much it sounds subjectively like a phone not at all a bit kinda completely How much it sounds subjectively like a phone Link back to matrix 99 a lot 100 Different Criteria One Can Adopt not at all a bit kinda a lot noise only completely not at all How much it sounds subjectively like a phone phone + noise Strict a bit kinda Decreasing d’ means decreasing the distance between the probability distributions. Note how, with a smaller d’, many misses and/or false alarms result, regardless of where the criterion is placed. phone + noise d’ (small) Probability Probability d’ (large) a bit kinda a lot How much it sounds subjectively like a phone 103 completely 102 Increasing d’ means increasing the distance betwen the probability distributions. Note how, with a greater d’, a criterion somewhere in the neutral area will produce very few misses or false alarms. not at all a lot How much it sounds subjectively like a phone 101 noise only Result: Pfa = .05 Ph = .50 d' = 1.64 Probability phone + noise Probability noise only Lax Different Criteria One Can Adopt Result: Pfa = .50 Ph = .95 d' = 1.64 completely noise only not at all a bit phone + noise kinda a lot How much it sounds subjectively like a phone 104 completely Questions terms of the SDT, what would be the • Ineffect of turning up the phone volume on the probability curves? What effect would this have on d’? What about turning down the shower? • A guard is watching the woods for visible signs of intruders. What are some likely sources of “noise” in his situation? 105 For More Info... A nice primer: Stanislaw, H., & Todorov, N. (1999). Calculation of signal detection theory measures. Behavioral Research Instruments, Methods and Computers, 31, 137-149. The bible of SDT: Macmillan, N. A., & Creelman, C. D. (2005). Detection Theory: A User's Guide (2nd ed.). Mahwah, N.J.: Lawrence Erlbaum Associates. 106
© Copyright 2026 Paperzz