Signal Detection Theory

[@fnymnky]

Here's a 10-min TED Talk I gave on rationality that illustrates how the Signal Detection framework can be useful in clarifying disagreements in everyday life: kzbin.info/www/bejne/pJiXm3WQrLB0m9k

[@gregorykarimian3813]

tried using this kind of logic with my gf and it still didnt work

[@daniellarios5426]

if you're studying for the mcat watch this video and not the khan academy one because that one is just all over the place. This one helped me understand the concept much better!

[@Jaylyons1]

Upping the speed to 1.25 makes this great.

[@shivakshsharma9377]

ma man 😂😂😂😂

[@CORPSE77]

Yo, why am I understanding this so good? thank you so much! I thought I would never understand this for the MCAT! Seriously, I wish I can have professors like this for every subject

[@mervei9668]

THIS IS SO GREAT EXPLAINED. THANK YOU!!! its these moments that make you realize HOW LUCKY WE ARE to have so many opportunities through the internet and that we can use it for both, "the good" and " the bad". just thanks:)

[@Jdiddy1792]

Such an easy topic, that my book makes seem so complicated! Thank you and congrats!

[@razzaxxe]

This is the best explanation of this I have come across. Thanks.

[@ishagulati248]

How can someone explain the concepts with such ease. Kindly consider posting more videos on various psychology based topics.😇

[@chutneypodientries3635]

Thank you for the amazing lecture, I truly wish to be amidst teachers like you.

[@kevuseth8027]

Way better than the Khan Academy video.

[@katherinecortez9337]

Simply amazing! Everything became clear when you provided the example! Thank you!

[@Arabella01]

Amazing explanation. Clear and straight to the point. Thank you!

[@rebeccashumway]

Thank you for posting! This helped me very much. You're a great prof.

[@hjjkthn]

what does y axis represent? is the frequency of a particular intensity(x axis parameter) like we have all y axis in normal distribution?

[@fnymnky]

Yes, that's right, these graphs are probability density functions.

[@kevinhernandez8283]

explained it better than my professor (and the useless TAs) at an "ivy league" (columbia university) did... my goodness thank you man

[@SonGoku-rl9qf]

wow, great job. Thank your for sharing this knowledge. You are a great teacher!

[@nachthimmel1998]

Thank you, finally I understood this concept :)

[@danjohncook]

Thanks Devin! Great stuff. Studying for an exam at the Berlin School of Mind & Brain, and your video was very helpful.

[@meredithebrown1676]

Thank you, made this a bit easier to understand for my Sensation and Perception course

[@pirtonian]

Thanks Devin, very easy to understand.

[@jimena6264]

Hi! I'm a student of Psychology in the Autonomous National University of Mexico, and I would like to ask you for the article that you mentioned at the end of the video about the problems for differentiating biases from discriminability, could you give me the exact reference of that article so I can read it or the website that you were talking about? Thank you very much

[@fnymnky]

+Jimena Arroyo I can't post pdf's here, and unfortunately it's not publicly available, but the paper is: Potts, A. J., Bennett, P. J., Kennedy, S. H., & Vaccarino, F. J. (1997). Depressive symptoms and alterations in sucrose taste perception: cognitive bias or a true change in sensitivity?. Canadian Journal of Experimental Psychology/Revue canadienne de psychologie expérimentale, 51(1), 57. Basically, previous research found that patients with depression had higher taste thresholds for sugar. People jumped to conclusions that depression had such fundamental and far-reaching effects as changing your very ability to taste, and started trying to link this to disordered eating and such. These authors came along and applied SDT to show that sensitivity (d'), the ability to distinguish sugar from not sugar, is equal across groups. All that changes is response bias, where depressed participants are more conservative. Cool stuff! Make sure you never just look correct trials!

[@anushabalasubramanian2068]

Thank you so much. This has been really helpful. I just had a quick clarification- If d-prime acounnts for the distance between means of the distribution, how is it different from standard deviation? What are the advantages of using d-prime over SD?

[@fnymnky]

+Anusha Balasubramanian d' can be thought of as a measure of the overlap between the two distributions, it depends both on the distance between the means and the variability of those two distributions (standard deviation). Two distribtutions can be relatively close together without much overlap if they have really small variance, leading to a large d'. The nice thing about d' prime as a measure is that these distributions are theoretical and typically not know by the researcher, but we can calculate d' based just on a participant's performance in a categorization task.

[@alexandra-stefaniamoloiu2431]

Thank you very much! Very clear and easy to follow explanation

[@nataliasteiner9430]

Thank you for this video, just one more question left: is there a formula or a special way to get to know d`, I mean how to calculate d`? Thanks for your answer.

[@JerryFrenchJr]

d'=2(B-C) according to other calculations I've seen. B = threshold, C = ideal observer. If you have a number line from -4 to +4 of intensity and the ideal observer picks 1.5 and your threshold for "yes" is 2 then 2(2-1.5)=1=d'.... and apparently this is called a C strategy which makes sense to maybe 3 people in the entire world.

[@fnymnky]

+Natalia Steiner Sorry to reply so late, you probably don't care anymore, but there's a great tutorial for calculating d' ("d-prime") here: www.linguistics.ucla.edu/faciliti/facilities/statistics/dprime.htm

[@sasali6727]

This link is a perfect explanation of D' and Beta. I do recommend it to everyone who tries to grasp the topic more in depth. Thanks Devin.

[@hosseinz]

Thanks Prof Burns, this helped a lot!

[@devinburns1186]

Hi everyone, thanks for watching! I want to encourage you to apply this conceptual framework to political conflicts around you. Many issues like food programs for the poor can be framed as a signal detection problem of determining how to evaluate applications. A food program false alarm is when someone is taking advantage of the system who doesn't really need help. A miss is when someone really needs and deserves help, but is denied. Rather than thinking that people are stupid who disagree with your position on shrinking or expanding such a program, a more charitable reading is that they are merely arguing for a different criterion. They may have a different perception about how to weigh the two kinds of mistakes and which one is more important to guard against. The hope is that if both sides agree that both mistakes happen and both are bad, perhaps they could work together to actually increase the discriminability of the evaluation process!

[@sallysueandyou]

that is an amazing thought! thank you so much for your concise video (and this comment in particular)

[@psychologic6216]

wow! such a compassionate view on differences in opinion! thank you so much for sharing this extermely valuable insight of yours!

[@PatrickRichardt]

This was very, very helpful. Thanks a lot!

[@jankicheese]

incredibly helpful, thank you!

[@navya120]

This was very clear and simple.. Thank you.. 😊

[@moniquejones5085]

Not sure why, but the audio is really now.

[@raymac6262]

Good summary, my friend. Thanks.

[@CMBiela]

Really appreciate the information! Khan academy MCAT prep covers this horribly!

[@janakaranathunga8476]

Nicely explained.! very helpful. thank you.

[@TheBF3kyle]

Student a Ulster University, UK using this piece for my dissertation research :P

[@thedeadman8361]

Excellent lecture! Thanks a lot.

[@copernicus99]

Hi Devin,Great tutorial once again! Might you have any videos or other material which cover how to calculate d-prime and do ROC analyses? Thanks

[@fnymnky]

Y Fish : I don’t, but there are many good tutorials out there. Somewhere lower in the comments I provided a link to a previous similar inquiry. Good luck!

[@torontogirl3]

thankyou for the clear and concise explanation!

[@xiaoxuanish]

Thank you. finally understood this theory

[@AT-qh3ug]

dang u explained it so well. helped me big time!

[@hoanguyenfutu]

Interesting. anyone who d love to share with me your experiences of applying this theory in your daily life?

[@fnymnky]

I can't stop seeing it all around me! As an example from my job, a colleague promoted a new tool as being better at identifying Alzheimer's, but never mentioned false alarms! It's easy to get more hits by shifting your criterion (bias), but you're not necessarily doing any "better", and could even be worse. I also see it in pretty much every political debate: those who want to restrict immigration spend all their time talking about potential "misses" (like a potential terrorist) while those who want to broaden it are concerned about "false alarms" (a hard working, good person who gets denied). It's pretty challenging to try to figure out how many of each case there could be and how much worse one error is than the other (in this case the terrorist allowed in would clearly be worse, but it's really hard to quantify by how much).

[@chutneypodientries3635]

Sir, do you know of any research papers where this theory has been applied in policy-making.?

[@maziarghorbani]

Congratulations. Thank u for the great video.

[@stepha.7638]

Wow!! Great Explanation!!

[@brentvanlieshout4207]

Hi, viewing this 10 years later and wondering... how is the kid?

[@fnymnky]

lol! She’s good: loves reading the Rick Riordan mythology books, being in plays, and playing Zelda with me 😊

[@rosedibya27rose14]

Best vedio are you an audiologist??

[@aanwar7381]

very well explained, thanks!

[@ruchisohal7554]

Simply amazing... #Thankyou

[@ishagulati248]

Hi Sir. I wanted to ask that is the bias determined by the individual himself (i.e. how accurate he wants to be before deciding upon a response hit/correct rejection) or is it influenced by external sources like the rewards and punishments one receives for giving that particular response or is it an amalgamation of both.

[@fnymnky]

It's definitely both. People tend to have their own default preferences, which can be related to risk aversion and their assessment of which error is "worse", but they can also consciously shift these priorities in response to incentives.

[@ishagulati248]

Thank you sir for resolving my doubt. Warm regards.

[@faresfares214]

thanks dude fantastic effort good luck

[@jl4260]

This was so helpful. Thank you! :)

[@copernicus99]

Great video. I'm a bit confused as to what the x-axis at 4:30 represents. Is it the 'objective' radar signal strength, or the subjective representation of the radar signal strength? Thanks.

[@fnymnky]

Y Fish : good question! It can be either, actually. In psychology we're pretty much always talking about the subjective experience of the observer. Even if we show them the exact same stimulus multiple times, sometimes they will perceive it as being stronger or weaker along whatever dimension we're judging. The same framework works fine for mechanical measurements as well though, like for a airport metal detector, for example. I guess you could still call this subjective in someway, as the exact same stimulus might be detected better or worse sometimes due to variability in the machine. I suppose the radar example really has two layers: the variability in the equipment, and the subjective interpretation of those signals by the operator. Does that help with your confusion?

[@copernicus99]

Thanks for the quick reply Devin! I'm still not fully comfortable in my understanding, but I suspect that this will come with time as I ruminate more on it. The most common label for the x-axis I've seen around is 'neural firing' which is subjective insofar as it reflects the observer's internal model of the external world.

[@fnymnky]

Y Fish - to flesh out your example, we could look at how certain neurons in the visual cortex distinguish between vertical and horizontal lines. Maybe for this neuron Vertical lines will tend produce a lot of excitation, while horizontal lines produce less. There will be some grey area in the middle though, with medium amounts of excitation occasionally coming from either orientation. To guess orientation based on firing, we have to pick a criterion or threshold, above which we will always declare "vertical". We will sometimes be wrong, with two different types of errors. To me, the biggest idea here is there moving the threshold higher or lower never makes you any better or worse overall: you just trade one type of error for the other.

[@copernicus99]

That makes sense. Thanks again Devin!

[@ahsinbeauty6438]

This was so helpful. Thank you

[@KenyaTeam2011]

Are the tables incorrect? Everything else I'm able to find states the Stimulus Present & Respond "Absent" is a FALSE ALARM. When the stimulus is absent and the response is "Present" is a MISS.

[@fnymnky]

Maybe you've seen charts with the axes reversed? If it's not there and you say it is, that's definitely a false alarm, and if it is there and you say it's not, that's a miss. These labels are more ambiguous when the task is discriminating between two things, like apples and oranges, but can still make sense if you frame it as an "apple detection task".

[@KenyaTeam2011]

@@fnymnky Thank you! I think I did have them mixed up-it was time to take a break in studying!

[@ashg8547]

This is great! Thanks!

[@jeffrey3690]

Nice summary thanks

[@claudia4578]

great vid

[@hjjkthn]

is bias the same as criterion?

[@fnymnky]

Pretty much. The criterion is the value on the x-axis where you put your decision bound, which will determine your bias (which response happens more often). Sometime bias is reported as the ratio of one response to the other.

[@e29g10h96]

Does anyone know how to create this kind of graphic with my own data in R?

[@devinburns1186]

I'd be happy to help! Unfortunately though, lots of data can't be graphed this way because the underlying distributions (e.g. signal strength/perceptual salience) are unobservable, we can only measure the responses. What kind of data do you have?

[@e29g10h96]

@@devinburns1186 Hello!! Thank you so much for your quick answer. I'm still new in this and still absorbing this new knowledge. I have some numbers for the four types of answer: hits (299), miss (210), cr(297), fa (125). Will this be enough data to create the graphic or do I need something else?

[@devinburns1186]

@@e29g10h96 With that data we can compute d'=.1, not super great :) If we assumed equal variability in the two distributions then we could make a graph, but that's not necessarily going to be the case. Here's my code: hitPercentage=hit/(hit+miss) faPercentage=fa/(fa+corRej) dPrime=pnorm(hitPercentage)-pnorm(faPercentage) x=seq(-3,3,.1) plot(x,dnorm(x,mean = -faPercentage),type="l") lines(x,dnorm(x,mean = hitPercentage),type="l")

[@e29g10h96]

@@devinburns1186 Thank you so much! Also, because of your last answer, I realised I got all the data wrong and the correct numbers (I think) (hit=3020, miss=740, cr=1290, fa=214). Thank you for your code and your help. I appreciate it so much :)

[@mcDynamit]

how do you get the distributions in the first place???

[@fnymnky]

The distributions are theoretical, there is usually no "ground truth" available for which stimuli come from which category. Also, we're often interested in perceptions of stimuli, and we can't directly measure those. In some cases you could build the actual distributions and quantify the real separability, but we usually just compute d' and beta from the response data without directly observing the distributions.

[@mcDynamit]

@@fnymnky thank you sir,, i was thinking maybe calibration of the instrument is useful for this,, for example if it is a radar we calibrate with birds,, and planes,,and get an idea of what each look like is that stupid thanks,,

[@fnymnky]

@@mcDynamit that’s totally right: if you can control or simulate the stimuli then you can better measure and calibrate the system.

[@mcDynamit]

@@fnymnky much appreciated

[@slena166]

that helped a lot !! thank you !!!

[@ranjanYadav-k3z]

can you pls provide me the slides?

[@fnymnky]

we could talk about that, you can email me at [e-mail]

[@anaodep]

Thanks :) Great video

[@medfac2010]

there is no mathematical developments . i need it .. please can you make a video showing us the use of math to resolve this prob

[@sugarxcookie27]

Thank you!!!

[@Moonehz]

thanks a lot

[@sherryzhou3158]

helpful，謝謝！

[@EndemikBitki-xc6li]

Gayet güzel:)

[@CockTailRapz]

fucking amazing. Best video in the net i saw regarding this topic. Big up!

[@taiman9423]

MCAT lets go

[@siam336]

I guess I'm an expert now

[@Jano-ws8nz]

f

[@raghavsingh4164]

Play at 1.25 speed