BIO FOR BUSINESSLAWLIB ARTS
BIO FOR BUSINESSLAWLIB ARTS BIO 301D
Popular in Course
Popular in Biology
This 39 page Class Notes was uploaded by Ezequiel Orn on Sunday September 6, 2015. The Class Notes belongs to BIO 301D at University of Texas at Austin taught by James Bull in Fall. Since its upload, it has received 7 views. For similar materials see /class/181708/bio-301d-university-of-texas-at-austin in Biology at University of Texas at Austin.
Reviews for BIO FOR BUSINESSLAWLIB ARTS
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 09/06/15
Friday August 24 2011 RationalRelevant Thinking o One year subscription of online access o One year print only o One year print online is same price as print only more people choose it o Without the middle choice people choose the cheaper option Money is made showing 3 options and people will want to choose the better deal Why Science Matters Why We Need a Method for Rational Decisions 1 Society needs top down protection from environmental contamination consumer protection setting priorities We need oversight to regulate what s going on we d be in bad shape otherwise 1a Technology impacts our health in indirect ways 1b Collective decisions are necessary o Cannot trust industry to regulate itself as individual we cannot prevent other individuals from impacting 2 Rational decisions by our government involve science o Issue Sciences gives us evidence amp conclusions o Other factors social political and ethical considerations AAA all of these make a POLICY Major historical example in which science was suppressed for ideology o Lysenko 0 Russian political figure USSR 0 Suppressed the science of genetics o US Stem cell research was suppressed on politicalethical grounds Monday August 29 2011 Continued from last time 3 We cannot rely on individual beliefs to reflect scientific truth o ie US public surveys show 0 Concept amp percent of individuals that believe Astrology 50 Witches 19 Aliens have landed 22 Communication with the dead 42 The above are class histograms of responses Main Points lots of support for things not supported by science lots with difference on some topics scattered some quirks bimodality rare and never extreme many patterns repeatable across years O 0000 4 How do we arrive at individual decisions Evidence 0 Experience of ours and others Value system Religionculture Experience Family beliefs what you ve been taught Views of others 0 Media friends Media World view compatibility Emotional appeal urban legends conspiracy theories Cost benefit analysis 0 People often believe what s in their best interest END MOTIVATION DISUCSSION How does the lamp mechanism work 0 cPressure oMetien O Hangman games gtexamples of the scientific method Properties and consequences of the scientific method 1 Goal understand the lamp mechanism phrase 2 Models guesses about how the lamp works 0 Your ideas of possible words in the phrase 3 Data observations made to test the models 0 Le lamp did not turn onoff in response to the flame o where letters went in various slots 4 Conclusionsevaluation deciding if model agrees with data thus far 5 Revision replacing old models as they fail with new ones 1 Science is NOT done as a poll issues NOT decided by majority politics or social merit Depends on evidence 2 Science does NOT accept explanations or models that lack supporting evidence We may start with a model that has no support but we won t keep it without evidence 3 Cyclical continual efforts to improve new data may always change your conclusions 4 Uncertainty is acceptable 0 Due to sufficient information Predicting the weather 0 Most accurate way to describe an outcome Coin flip reflects an understanding of the issue 5 Progress as a consequence we keep going to reiterations of it step 3 we aren t guaranteed progress but it often happens 0 Goal is driver all progress is measured relative to the goal 0 Various examples of progress over the years see book Smoking a No concern gt lung cancer 1950 gt heart disease gt secondhand smoke gt effects on fetus gt nicotine addiction Institutions that lack one or more elements of the scientific method Religion No pretense of trying to use the scientific method Astrology 0 has goals 0 has models rules for making predictions no attempt to evaluate these rules and modify them accordingly Government agencies 0 goals social improvement of some sort 0 models bylaws describing how the agency operates o revision changing bylaws typically weak on data and evaluation or the explicit purpose of testing the goal Criminal trial 0 goal determine innocence or guilt 0 model trial model of crime model 1 defendant did not commit crime model2 defendant did commit crime 0 data evidence in trial 0 evaluation verdict o revision weak appeal reconsideration of verdict any time new data obtained History of failures to heed evidence use SM o Lysenko ideology something about Soviet genetics o Copernicus and followers catholic church opposed their discoveries about the solar system church recounted 300 years later Now evolution faces opposition from some religions o Benefit of physician hygiene 1818 1865 o Discovered that women had a high rate of death 10 of women in maternity wards was due to physician s dirty hands o South Africa and HIV as the cause of AIDS 19992008 0 O Mbeki government official stance opposed this view It s common in politics to neglect evidence stick to a plan regardless of outcome Kennedy administration actually used the SM Bay of Pigs Teaching methods END OF THE SCIENTIFIC METHOD Frida Se tember2 2011 Model anything idea plant thing organism that we use to represent something else as a shortcut to achieving a goal o Model is judged by the goal o Examples of models model organisms in biology O 0 00000 0000000 bacterial viruses foundations of molecular genetics DNA is genetic material bacteria many of molecular genetics used as screens for cancercausing compounds yeast identifying genes that control cell division flies xrays cause genetic damage worms programmed cell death fish developmental genetics mice 99 of genes in humans have counterparts in mice 30000 genes total NIHgoal of creating knockouts for most mouse genes Showed scale model of solar system brand name model of the company and quality pictures models of objects titles models of contents past model of the future reaction of others model of our reaction slogans 5 points about models o 1 All models are false 0 Every model has imperfections in some way 0 Not exactly what it represents 0 No such thing as a perfect model 0 We use a model when its limitations don t matter 0 We quit using a model when its limitations do matter o 2 Usefulness of a model depends on multiple characteristics 0 ACU o A Accuracy similarity to what it represents 0 C Convenience how easy it is to use cost ethics 0 U Uniformity how similar one copy of model is to the next 0 Consider models of humans for genetic purposes rank Model a Human 1 least convenient n Monkey 2 n Mice 3 n Yeast 4 n Bacteria 5 most convenient Usefulness of a model is measured relative to the goal o 3 3 types of models 0 Abstract ideas theories math computer numbers graphs 0 Physical touchable organisms structures demonstratives 0 Sampling how study subjects are chosen and divided UP o 4 Onetomany manytoone no such thing as just one model of something there are uncountable numbers of models of anything o 5 Pieces and parts as model 0 A model of something can be a small part of that something Wednesday September 7I 2011 5 issues to address with any model o 1 Goal o 2 What model is o 3 What model represents o 4 How model might be useful for the goal o 5 What are the limitations of the modelfor the goal Condom testing Why use condoms Goal Act as a barrier to o 1 Avoid pregnancy 12 failure ratecouple o 2 Avoid STD s 12 million new STD s transmissionsyear in the US Sales in US 1 billion condoms2 years Why would condoms ever fail ie breakage failure Goal Maintain sensitivity 0 Conflicting goals in use and manufacture of condoms Barrier Maintain sensitivity We test condoms to maintain standards of quality In the US failure rate of 51000 means that the batch being tested cannot be sold 0 Not an absolute standards of zero failure rate How might we test condoms 0 Goal ensure that condoms don t break during use Models and possible models for condom testing 0 ACU Most accurate model possible 0 1 Trained sex technicians A U o 2 Volunteers A sort of C There are no government mandated tests of condoms that involve humans 3 Mechanical tests C U Real advantage for industry because they can engineer condoms to pass mechanical tests as long as the test is uniformly applied Multiple tests applied to all condoms some tests applied to samples ie nondestructive electrical test applied to all condoms before packaging would pick up holes Destructive sampling of a few condoms in a batch using a physical endurance test Water leak test fill condoms w 10 ounces of water look for leaks Stretch test cut a ring of condom and stretch to its limit Airburst test fill a condom with standard volume of air and see if it breaks 0 Demo Validation tests have been inconsistent in supporting airburst test as a good predictor of breakage during sex 0 O O But choice of which model to use when testing condoms is NOT based solely on breakage during use Other limitations of airburst test and other mechanical tests 0 Do they address STD passage NO 0 Water leak test that supposedly identifies holes of 5 microns or larger Showed scale model of 5 micron circle relative to various STD agents all agents could pass through 5 micron hole Tests of HIV transmission using volunteers 0 Discordant couples one is HIV one is HIV and you observe the rate of conversion infection of the HIV individual to HIV Cannot get strict adherence to condom use Several studies consistent use vs inconsistent use 0 0200 101200 Consistent condom use decreases infection rate Friday September 9 2011 See new table on website on condom testing chapter Models and DWI testing Stats 1824 year olds 8000 deaths on highways annually 0 12 are alcohol related 0 For college students 15 deaths 100000 per year Approach here and in most other countries is deterrence penalties for drunk now impaired driving Models if impairment 0 What is relevant to driving Vision Judgment Coordination Reaction time 0 Ideal model of impairment would asses these and all of these Texas laws legally impaired when either or both 0 Not having normal use of physical OR mental faculties 0 Have a blood alcohol concentration BAC of 008 or more Older laws in some states started at 15 then 12 then 1 now 08 Legal ways of getting a BAC breath urine and blood What about Standard field sobriety test aka SFST o 3 components A Walk and turn WAT B One leg stand OLS C Horizontal gaze nystagmus HGN n Involuntary eve movement A amp B test coordination C tests o What about testing your mental faculties Ability to follow directions Thirdconvenience issues would be a road test Back calculating a BAC o If you get stopped sometime later blow under 08 the legal system may still go after you 0 There are ways to back calculate a BAC at the time you were stopped 0 These data from ideal situations single shot of alcohol empty stomach they don t look so clean with food in stomach and alcohol consumption over time Monday September 12 2011 Limitations of models in DWI testing o If goal is to identify impaired drivers limitations of models are 0 08 BAC using blood 08 doesn t cause same level of impairment in all 0 08 BAC using breath Same limitation as above but also breath concentration may not reflect the blood 0 SFST scoring of the test is somewhat subjective performance may be affected by conditions other than driving ability 0 Back calculation effects of many variables on rate of BAC decay over time not known END DWI TESTING Infectious diseases models in epidemiology o Some new infectious diseases 0 Mad cow bacteria 0 Lyme 0 Flu H1N1 bird Him 0 HIV 0 West nile o 2 fear issues 0 How bad will it be ifI get it o How likelyI am to get it possible health concern o One goal reduce cases prevent spread o How many infections diseases have been eradicated by humans Small pox SARS o No matter how bad a disease is if we can prevent it from spreading in the population it will die out assuming no animal reservoirs 0 Even if we can t eradicate it reducing its spread means fewer people get infected Goal of public health 0 Some simple concepts and math underline powerful principles R basic reproductive number of a disease a Average number of new infections started by the first infected individual in a population kind of birth rate Initial spread of a disease is like a chain reaction a 1 gt Rgt RAZ gt RA3 What value does R need to exceed for disease to grow in population If R gt 1 disease becomes epidemic and grow in population If we can make R lt 1 it ll die out Influenza 153 Measles 518 Chicken pox 712 Polio 57 Small pox 1520 HIV 212 SARS crowded 2236 o SARS community12 o Lots of variation even within a disease Why 0 R is a limited model not the same across all environments 0 R 1 Rlt 1 considered epidemic threshold o RBSrd o S number of susceptible hosts 0 B infection rate parameter how easily disease is transmitted r rate of recovery of infected individuals d rate of dying of infected individuals B r d are all properties of infected individuals 0 S is a property of the population o The formula tells us how to achieve R lt 1 0000000 000 Wednesday September 14 2011 S susceptible B infection rate parameter r recovery rate d death rate of infected host Masks and environmental cleanup do what o Affects B decreases R decreases A drug that hastens recovery does what o r increases R decreases Vaccination does what o S decreases R decreases We can calculate fraction of population needed to be vaccinated to eradicate disease o Write S after vaccination as xS o x is fraction of population susceptible after vaccination 0 New R XBSrd xRod lt 1 o X lt lR We have vaccinated 1x or 11R to eradicate If R prior to vaccination is 3 we need to vaccinate 1 13 23 of population to eradicate o if R was 30 we would need to vaccinate 1130 97 of population Most obvious result need higher levels of vaccination for diseases w higher R Since most vaccines are imperfect not everyone who is vaccinated is immune it may be impossible to eradicate a disease w a higher R Model strengths o Tells us what to change to reduce infection levels in population 0 Vaccinations 0 Reduce transmission 0 Speeding recovery Model limitations o Assumes mass action homogeneous population and environment o Assumes constant transmission B in space and time o Not fit STD s or animalhuman transmission NEW CHAPTER Extrapolating Health Risks o Thousands of chemicals manufactured used by industry were dumped air water ground products o What does exposure mean for us Long term and short term effects of these chemicals o How to figure out 0 Test on animals 0 Doses issue o Bottom line is we extrapolate type of model 0 1 Across doses 2 Across species 3 Across effects 4 Across related hazardschemicals Elaborate 1 Extrapolate across doses n Easiest to study effects at high doses o Things are easiest amp fastest to measure at high doses getting drunk D But we potentially care about long term effects and subtle effects those that would follow from low doses n Presents a backcalculation problem c We see an effect at high dose how do we calculate an effect at low dose a Consider 4 types of extrapolations from high to low 2 Extrapolations across animal species a Rodents are used as models of humans for most tests of carcinogens cancercausing agents D But what kills a rat might not kill a human at the same dose 3 Extrapolations across effects a May use a study of toxicity what kills to extrapolate to a cancer risk 4 Extrapolation across related hazards a Way to study health effect of one chemical and use it to represent all forms of that chemical especially radiation 0 O O O FridayI September 16I 2011 Extrapolations models of health risks 5 examples o 1 In the 1990 s the US set a standard for exposure limit for acceptable exposure limit as 11000000 x LD50 in a suitable rodent model 0 LD50 dose per gram body weight that kills half those exposed o No rationale except that it seems conservatively safe 0 Linear extrapolation backwards o 2 Rodent models of cancercausing agents 0 Problem in dose extrapolation o How done give highest exposure possible look for tumors in rodents o Objections raised by AMES Doses chosen for testing nearly toxic Toxicity leads to cell proliferation hence cancertumors tumor elevation is an artifact of this procedure Rodent model affected by an accelerating threshold dose response o 3 FAS Fetal Alcohol Syndrome Drinking pregnancy can lead to problems for fetus Initial data suggested a problem w heavy drinkers Advice 2030 years ago was that 12 drinksday was probably okay 0 This advice assumes an acceleratingthreshold model no evidence to support that o More recent data suggests effects at even 1 drinkweek ie extrapolations replaced by data o 4 Secondhand smoke 0 NOT an issue 23 decades ago although linear extrapolations would have indicated a risk direct measures are now possible 0 Direct measures now possiblelinear model supported 3000 deathsyear attributed to secondhand smoke o 5 Dioxin byproduct of a herbicide maybe most expensive chemical hazard in history 0 Agent Orange love canal 0 Early studies suggested dioxin was HIGHLY toxic 1ppb lethal 0 Lots of money spent to clean it up 0 Toxicity tests based on guinea pigs which we know are REALLY REALLY sensitive to dioxin Humans rats mice not nearly so sensitive 1000xlen All these examples demonstrate how messy our understanding amp models of health risks are Radiation o Lots of beneficial uses of radiation o Common misunderstandings 0 Effects almost universally harmful NOT teenage mutant ninja turtles Real effects genetic damage cancer possible birth defects mental retardation of the fetus O O O o How much causes harm low increase don t have much effect on cancer rates 0 Difference between an nuclear bomb and pollution or reactor meltdown Themes physical models in radiation 1 Organisms humans only o Can use other organisms to understand that large doses are harmful o But for doses of interest only humans are accurate enough to use 2 Effects of radiation what models do we study o Humans get lots of cancer but most are too rare to study in this regard o Most convenient cancer is leukemia 0 Background rate is high for cancer 0 Short lag time from exposure to cancer 5 years 3 Types of radiation ionizing radiation high energy o 2 fundamental types 0 Electromagnetic ultraviolet xrays gamma rays cosmic rays 0 Free atomic particles neutrons electrons B beta alpha particle NEW SECTION Monday September 26 2011 Data our connection to nature 3 critical questions to ask of any claim o What s the evidence aka data o How were they obtained This section of the course o Who obtained it Last 2 weeks of class Data specific type of model o All models falsesome true of date Hereall data measured with error or possibility of error Error does not mean a blunder o Mean that data often differ from what they represent perhaps slightly Protocol procedure o Model of how the data are collected 0 Influences the types and magnitudes of error 0 Allows the data to be gathered consistently o Allows someone who did not gather the data to look for types of errors that might be present Protocols for gathering data o Protocols for analyzing data o Protocols for reporting data Several problems can arise in data Recognize 4 types of error o Coin flip tally most of us did not get expected 5 heads in 10 flips o 1 Sampling Error Games of chance Diseasecancer incidence Sons vs daughters people family size attendance at game store Accident rates Also attribute chance to many events in our lives when we could not have predicted them 0 Fixing Sampling Error 1 We can reduce its impact magnitude by taking a larger sample 2 Taking a larger sample gets us closer to true values 3 We say the fix for sampling error is replication multiple sample 4 A hierarchy of sampling error and replication n Lots of ways to replicate a study a Say we take 100 of you in this class calculate an average SAT score Is there any form of replication in this sample i Yes we have 100 different student SAT scores Only if we had just 1 student would there be no replication Is replication absent in some fashion in this sample 1 Always absent when you have sampled everyoneeverything that could address your question How much replication is enough 1 It depends on the goal and how accurate we need to be 2 For public health issues we may need a trial of 100000 to be safe 3 For many other purposes small sample might suffice o 2 Rounding precision and accuracy Width of dime 1785 mm Exact No exact measure would go out many many decimals RPA limitations in our equipment and general abilities to make quantitative measurements a Those that involve decimals speed weight time distance Only fix for RPA error better equipment 0 3 Human and technical Mistakes and blunders Machines not calibrated correctly Technically avoidable in a strict sense but can always happen Fixing human and technical error a Design protocols to reduce human and technical error 0 4 Bias unintentional Wednesday September 28 2011 Human and Technical Error o Fixes o A better protocol 0 Standards measure an object of known value ie for a thermometer a standard would be boiling water ice water body temperature of healthy person 0 Independent replication A proficiency test is use of a standard to detect H amp T error how well something performs by using a standard o Comment standards and proficiency test may NOT fix H amp T error but they allow the determination of an error rate o Reference database similar to a standard Unintentional Bias o Think of an odd random number 0 7 is the number chosen the most often and amp 3 is the second most shows that this is not random and when people are asked to choose things randomly they most often do not o Bias is a consistent error so replication increasingly gives us the wrong answer does NOT fix bias o Fixing bias 0 If choosing choose randomly from a hat roll the dice random number table 0 Blind do things take observations without knowing what is expected Two levels blind subjects only needed if subjects are people subjects are the individuals in the study a Blinded subjects means they do not know their treatment status or what is expected of them a EX in testing a medicine how do we blind subjects so they don t know whether they got the medicine or not Give them a placebo fake pill etc Blind observers those who make the measurements do not know the treatment status of the subjects whom they measure If both subjects and observers apply Double blind Revisit Protocol o Protocols specific level of replicity use of blind and randomization standards level of precision and accuracy equipment and procedure that would influence HampT error o Now common in US bureaucracy to evaluate adherence to protocol as a substitute for data quality o Diagram of where errors arise 0 Nature choose things to be measured and measure them Affected by n Bias sampling error RPA HampT error some bias Friday September 30 2011 Data quality context the Criminal Justice System forensics DNA Typing o You re on a jury for a serious crime There is a suspect almost no evidence except blood from the crime scene and it s not the victim s blood Lab has declared a match between the suspect and the sample o Step 1 0 Match is based on blood type both are A o Willing to assume suspect is the source of the sample 0 No chance of accidental random match is too high o Step 2 0 Second analysis is done 0 We are told that the random match probability RMP is A 1 B 01 C 001 D10 A5 E 10 A6 F No threshold At what point do you agree that the sample came from the suspect As RMP goes down more and more of us willing to assume suspect is the source o Why do we care about DNA typing 0 Gives us very low RMP s o Where can we get your DNA 0 Blood Semen Saliva Skin cells Hair 0 Poop o How can we work with tiny quantities of DNA 0 DNA Xeroxing PCR 0 DNA good for solving crimes because Each of us has a single profile same in all cells Low RMP s possible and can calculate them easily DNA is easily left behind in usable quantities History of DNA Typing o Now 3 different sources of DNA used in typing o A STR short tandem repeat 0 Most informative of the three methods lowest RMP 1 in 10100 million 0 Requires the highest quality DNA not always usable 0 DNA type consists of around 20 numbers o B Mitochondrial sequence 0 Not terribly informative RMP 110 11000 0 People with the same mother maternal grandmother have the same mitochondrial type 0 Easy to get mtDNa even from degraded samples C YSTR o For typing males based on the Y chromosome 0 Not terribly informative but gives an estimate of number of males in the sample o Exclusion principle if any part of a DNA profile does not match between sample and person they do NOT match o Protocol in DNA forensics is o 1 Obtain samples from the crime type the DNA 0 2 Obtain samples from suspects type the DNA 0 O O O o 3 Compare the DNA types 0 4 If a match calculate the RMP need reference database 0 5 Decide what the match means was suspect really there What makes typing Properties needed DNA method reliable RMP attainablereliable Reference database Match is all or none Characteristics are discreet Independent Protocols are universal verification is possible Assurance of low HampT Labs pass blind error proficiency tests Monday October 3 2011 Science and US Criminal Justice System A revolution from DNA typing Some context Innocence Project As of 2011 273 convicted people in US prisons have been released after new evidence showed they could not have committed the crime at least 15 were on death row As of 20012002 67 people had been freed by DNA tests done after conviction at the time this was 23 of those tested Other side In 2003 Houston Crime lost its accreditation for DNA typing because of sloppy procedure Most wrongful can be attributed to bad data data analysis when identifyingmatching people or forensic objects Besides DNA what other methods do we have of identifying people or matching samples o Fingerprints o Hair not DNA based o Dog smell o Shoe print o Bite marks o Eyewitness identification o Bullet marks bullet lead analysis How is data gathering and analysis used or relevant to forensics o Gather crime scene samples o Search for other evidence o Interview witnesses o Process samples o Interpret and analyze data Type of Error Where it can Consequences Examples arise HampT All tests False Sample mixup recording and incrimination storing false evidence exoneration analysis Bias Interpretation of Data can be Eyewitness subjective made to fit a interviews lab evidence choice suspect procedures not of evidence blind falsification of evidence RPA Some types of Matching criteria Bullet lead lab analysis can be come analysis video subjective Sampling Occasional Misinterpreted Recent match of fingerprint suspect and match sample by chance Ideal features for a forensic matching procedure Feature Why needed Errorreduction Flagsindicators principle of absence Reference Allows A form of a Not mentioned a database calculating standard to claims that a chance of an accidental or coincidental match RMP calculate rate of a wrongful match match is unique Characteristics measured are Clear whether a sample has it or No RPA error No description w specific discrete not allowing characteristics consistent that define the scoring match Independent Someone else Form of Methods of one unification of can verify or replication to expert cannot be match is challenge the detect many evaluated by possible conclusions types of error another no universal explicit protocol protocol characters permanent Labs are subjective to and pass blind proficiency test Provides Form of No error rate assurance of standard to given tests method estimate an internal NOT accuracy overall error blind rate undocumented Wednesday October 5 2011 Review ideal forensic method 1 Reference database 2 Discrete characteristics 3 Independent verification 4 Pass proficiency test Last time We watched a bullet lead analysis video 1 Reference database No nobody else in the country was able to do the analysis It s ambiguous Probably not Fingerp 2 Discrete characteristics No not looking at them red flag 3 Independent verification No nobody else was doing it nobody knew what the protocol protocol was didn t even really have 4 Pass proficiency test No someone came in and said they were doing everything all wrong Dog Sniffing 1 Reference database No couldn t use it anyway not effectively 2 Discrete characteristics No odor isn t done that way 3 Independent verification No protocol Can t tell a dog what to do or smell characteristics are not permanent 4 Pass blind proficiency test No when judge set up test and dog failed that means it isn t blind anymore rints gold standard before DNA In the 1990 s voluntary proficiency tests found error rates of 10 o20 up until then everyo ne assumed they were perfect Before 1990 After 200 1 Reference Database Yes but could not be used for RMP Yes proper one 3 Independent verification Yes No no universal protocols 2 Discrete characters No Yes refused to accept those as a way of typing 4 Proficiency tests No Hair matching used until about 2000 Proficiency test in 1970 s found error rates of 2868 Eyewitness identification o Reference database Sort of but not used properly to get RMP o Discreet characteristics NOT the way people we recognize faces o Independent verification Not possible to specify a protocol for what people remember memories are dynamic so what is remembered is not permanent but often have multiple witness identify same person o Pass proficiency test Instructions make a huge difference we saw the lineup and picked someone but it ended up that he wasn t there 100 years of tests showing how bad we are Monday October 10 2011 RMP 11 million LER 1100 2 ways that suspect unequal sample accidental match what do I do with this evidence What is the chance that this suspect is not the source of the sample given the information how do you put these numbers together given the chance that the suspect does not equal the sample Add them 01000001 WHAT TO STUDY Graphs why difference in shape of graphs Go back to previous years exams he s borrowed exact questions Graph thing Combining Error Rates o Where do numbers come from Four ways to obtain a count o 1 An actual total count o 2 A simple extrapolation o 3 A simple compound extrapolation o 4 Fictitious made up 1 Actual counts attempts to include all date o Athletic statistics o Census records c Subject to errors of course 0 inadequate protocol 0 HampT error 2 Simple extrapolations o Record data on a subset w whole o Apply that number to the total o Polls and any other method that gets the specific data we want on a subset of the population of interest o 2 additional types of errors 0 Bias 0 Sampling error 0 In addition to those for 1 3 Compound extrapolation o Gather some data convert the data before using c Common in medicine and public health to report incidences of various diseases But diagnoses of what people have usually very indirect Also subject to error above especially bias and inadequate protocol o 1964 Gulf of Tonkin Resolution based on sonar data misinterpreted it and thought we were under attack 4 Fabrication o Media often makes up numbers to suit themselves o Numbers get quoted by other sources and eventually become fact Monday October 17 2011 Article about FBI agent Null hypotheses article of dietary supplements o Makes a big difference which null hypothesis you start from with FDA approval and drugs if null hypothesis is safe until proven harmful NOT much expense on your part but if it s harmful until proven safe big expense Falsifiability of models dog article pet psychics pet communicator o Range of difficultyease in falsifying models o Consider models for the probability of heads in a coin flip o Models P 12 P 13 P34 Very falsifiable because we can easily imagine data that would falsify them or alternatively support them P lt 12 still falsifiable but more difficult than AAAAA P lt 12 or P gt 23 still falsifiable but more difficult than AAAA o Model P not equal 12 only rejectfalsify only if P exactly V2 o Can t do would be the same as proving P 12 No amount of data can do that General point models vary in how falsifiable they are for any set of data there are always countless models that are consistent so we cannot prove a model is correct Because it would be difficult to try and reject countless alternative models we use null model approach to avoid dealing with alternatives that seem implausible No theory can explain everything To be accepted a model must be 1 be supported and explain some observations and 2 is not related by others Ie astrology may be too vague to be refuted but since it does no better than random in tests we don t hold onto it Criteria for rejection Consider rolling a die 10x All of them come up with 6 Null model is all faces have an equal chance of coming up Could we get 1010 6 under our null model 0 YES How then do we ever reject our own null model 1 In this case we would use statistics to calculate likelihood of observations all short term 0 If this likelihood is small we reject model 0 There is a common rule in science that if probability of results is less than 005 we reject 0 NOT hard and fast 05 means that you will wrongly reject a correct model 1 in 20 times 2 Non broadly in science we used rule of repeatability if multiple observations over time are NOT consistent w a model then we eventually reject it long tem Let s discuss the following pattern Red cars are involved in more accidentscar than are cars of other color do not believe this pattern when you leave class correlation Why might such a pattern occur causation o 1 Accident prone drivers prefer red 0 Red cars harder to see 0 Types of cars painted red are crappy 0 Red easier target Why tells us a reason for the correlation 0 But we cannot go backwards cannot easily argue a cause from a correlation Why refers to What is a correlation It means that 2 things varychange together on average An association between 2 variables if you know value of 1 variable you know something about the value of the other variable Ex in biology correlation between driver age or gender and car accident rate Age and all sorts of health issues Gender and height Gender and hair length Family size and income Season and temperature SAT scores and university grades OOOOOO Wednesday October 17 2011 correlations Graphic 0 al approach to correlation Positive slope with points along line positive strong Negative slope with points along line negative strong Positive horizontaly slope with dots around it positive weak See book easy way to go from a verbal description to deciding if there is a correlation is to GRAPH or imagine it The data need 2 variables and one must change on average as the other changes Correlation versus Causation What s the difference 0 Correlation is a broad statistical pattern 0 Causation gives us a reason for a broad statistical pattern We care about causation it tells us what will happen to y if we actively changed x Exs of the difference between the 2 o Causation Smoking causes lung cancer causation tells us that if we smoke our chances of lung cancer increases Correlation Smokers have higher rate of lung cancer than nonsmokers Causation Talking on cell phones leads drivers to have higher accident rates Correlation Drivers talking on cell phones have higher accident rates than drivers not talking on cell phones Causation Studying improves exam scores 0 0 Correlation People who study have higher exam scores than those who don t study We want casual explanations they tell us how to achieve our goals BIG THEME Right here Correlation DOES NOT cause imply causation DO NOT ASSUME THAT 2 VARIABLES IN A CORRELATION ARE CAUSUALLY RELATED Reason Correlations are consistent with many causal models Why we need to pay attention our brains just usually jump to conclusion that correlation is causal Advertisers take advantage of this o Ads for many products simply try to develop an association between the product and good feelings o Examples of mistake correlations o Malaria infections disease malaria means bad air thought originally to be from breathing unhealthy air 0 TB same AIDS first guesses were that it was caused by drugs Lots of mistaken inferences in diet and health because most evidence is correlation 0 Sometimes a correlation is due to causation though Cowpox is a vaccine against small pox Smoking and lung cancer etc News articles commony infer causation from correlation Why can we not correctly infer causation from correlation o IfX is correlated w Y have 3 points caused models 00 o I X causes Y o II Y causes x o III A 3rd variable 2 causes X and Y figure out what Z is Represents what we re going to call the hidden variable problem Revisit red card have higher accident rates Raises some obvious questions about causations such 0 00 as If YOU buy a red car will your accident rate go up IF we banned red cars would accident rates go down 0 To decide if red is causal we need to know if changing car color alone would change accident rate Causal models Friday October 21 2011 Higher Accident Rate with Red Car Causal IWhyRed IWh ICausal I339dHidden Will Y Model Cars Other Variable Variable Accident Dangerous Color Rate Cars Change if Safe Color Changed Dangerous Type of Safe Type of Yes No cars tend Car Types type car to be red often red of car are dangerous Red color Unsafe Safe Color No hard to color color see People Red elicits Color No Yes become aggression more aggressive when driving red Note In this last model red is still considered the cause even though it is caused only through a chain of effects Still on our correlation that red cars have higher accident rates Hidden variable problems look like this o Red cars risky drivers sports cars old cars nonred cars safe drivers safe cars young cars The third variable problem would go away if missing notes here damn computer Suppose we have data as follows A red sports car B red safe cars C Nonred sports cars D Nonred safe cars Car color Red and Other Sports High Safe low To simultaneously eliminate types of car and types of driver as variables we would need a 2 x 2 x 2 table A more quantitative version of the same thing is Sp9rts Safe Red 5 L Not Red 10 2 Accidents per 1000 cars year If most red cars are sports cars and most nonred cars are safe cars then red cars can have the higher accident rate 2 Don t eliminate it but you spread it out equally over treatment and control groups When they are evened out like this they can no longer be the cause of a correlation between color and accident rate Goal is to control for as many factors as you can Challenge in doing so is o i We can only attempt to control for factors we can think of o ii Even then it may be difficult to get complete data 0 MondayI October 24K 2011 2 models of intelligence brain size and IQ score Progression of tests with better and better controls Absolute brain size relative brain size controls for body size IQ tests what variables were controlled for o Language c Test environment Penguins Humans Phase 1 InteHigence X Inte gence Y Unfamiliar Familiar environment environment unwanted variable that may lower penguin score Unfamiliar language Familiar language unwanted variable that may lower penguin score Phase 2 Controlled for environment Phase 3 Controlled for language General problem is correlations may NOT reflect causation because of 3rd variables o Solution Control for these 3rd variables Two general approaches to controls o A Sort through the data that nature has provided identify 3rd variables and use treatment to control groups matched for those variables o B Gather a new type of data do an experiment ie you gather data where nature has been altered so that the 3rd variable is controlled 0 Experiment a deliberate change in the natural order to test a model manipulation In A we are taking data as they exist and sorting them In B we are altering the way data are created Two types of experiment o i when a specific 3rd variable has been identified you gather data in a new setting where that variable is eliminated 0 Ex You feel jittery after drinking sweet coffee in the morning is it the sugar or coffee Drink coffee without sugar or have the sugar without the coffee 0 Ex In the video we just saw eliminated test environment as a variable by giving the test to penguins in the same environment as humans o ii Randomly assign to treatment and control group 0 3rd variables are not eliminated they are evened out between control and treatment 0 In a drug trial you would randomly assign who gets pill vs placebo If you let people choose what group they get let into would not control for 3rd variable 0 In one car color red vs others what would constitute an experiment that controlled for all hidden variables but color Magically change color of existing cars randomly not feasible of course Now the drawback of A using correlational data is you may never guess what the relevant 3rd variables are Then you could never control for them This limitation also applies to the first time of experiment o Drawback of experiment Wednesday October 26 2011 Watched primetime autism video Child saying heshe was sexually abused Studies showed it was the facilitator who was talking not the child o Unconscioussubconsciously controlling behavior FC Experiments facilitated communication Model 1 words for child Model 2 words from facilitator Normal FC environment confounds the facilitator response with the child response Need an experiment to destroy this correlation Explicit protocol Obviously yes Blind yes neither the child nor facilitator knew what the other was seeing Replication Yes multiple facilitatorchild pairs and multiple trials multiple institutions multiple types of tests Controls Yes When shown the same picture Randomization Ambiguous could have randomized order of the pictures but we don t know if they did Friday October 28 2011 Experiments with humans Clinical Trials experiments with people o Purpose substance safety and efficacy 3 Phases o Phase I 2080 people 0 Toxicity o Phase II in the hundreds 0 Preliminary efficacy o Phase II several thousands 0 Final test of efficacy o Phase IV extending approved treatment with a drug already approved for other persons Main Points of the Video It s indeed possible to apply the scientific method to psychic phenomena When psychic predictions are vague or numerous we cannot refute all of them can t even hope to We use a nullmodel approach to ask if psychic predictions do better than random or better than some other model that answers on special powers a Horoscope Model people s responses reflect the accuracy of the horoscope Manipulation everyone got same horoscope Replication yes 30 students Randomization NOT relevant if all students got same horoscope Blind yes student unaware all were same Control No or ambiguous we assume that class represented many astrological signs if all Pisces and all horoscopes were for Pisces then we would expect high scores under the alternative model B Palm reading Model responses reflect accuracy of descriptions Manipulation told clients the opposite Randomization no Control yes the people whose palms were read correctly Replication yes pretty minimal Blind Present at all in some fashion One sided client did not know but the reader did Explicit protocol for both sets of experiments Monday October 31 2011 This stuff isn t in the book Experiments do not eliminate all possible 3rd variables they just eliminate those within a study o Assumes the experiment is well done 1 Study health effects of smoking o Treatment Salem filter cigarettes o Control No smoking o Variables NOT included other brands nonfilters other years c Point is any observed effect of smoking in this study could be specific to Salem brand the filters or to those years 2 Study effects of drug X Treatment Drug X given to 2030 year old males Control No drug given to those males Variables NOT included other ages females other years Point is any effect you observe or lack of could be specific to 2030 year old males etc o A study cannot control for variables that lie outside of its domain TEST QUESTIONS Texting while driving increases a driver s car accident rate describes causation so it s WRONG People who talk on cell phones while driving have higher accident rates than people who do not talk on cell phones while driving YES it s a correlation The UT tower is orange on nights of a major athletic victory but is not orange on nights of a major athletic loss it s just a description of a pattern YES it IS slope isn t zero Wednesday November 2 2011 New section Impediments Why the scientific method might not work why we aren t using it Problems with implementing the scientific method Today intrinsic difficulties problems when nature is stacked against us 1 Humans are difficult research subjects All sorts of issues ethics cost behavioral compliance lack of suitable model systems Ethics there is a lot we aren t willing to do to ourselves o In west now strict government protocols to ensure the safety and informalness of human subjects 0 Review committees consisting of uninvolved researches to avoid conflict of interest Sometimes concern about subject wellbeing backfires 1St major clinical trial of AZT abandoned the control group used for AIDS Partway through the study the researchers were so confident of AZT success against HIV that they started giving the drug to everyone in the trial but when the trial ended they realized the effect was so small that they couldn t see anything Cost major factor in some human research the overall cost of getting new drug approved by the FDA runs anywhere from 5 billion dollars to 1 billion dollars Compliance are people in the study behaving as they should or as they say they are Big issue in diet and health Above problems are exacerbated made worse by the fact that there is often no suitable animal model system to study what we want to know 2 Rare events are difficult to quantify relative concept Imagine a slot machine that s supposed to pay a million polls Would need at least 10 million polls to be able to tell statistically that is it not living up to the advertised win rate etc Rare events pose 2 problems 0 1 That you need to detect the event 0 2 Overcome sampling error to know the approximate frequency Rare can be on an individual basis or a population basis Couple of examples 0 A Clinical trials phase 3 involve a few thousand patients they will fail to identify problems that will occur less often 110000 or more rarely once marketed there may be millions taking the drug whence rare side effects will be discovered with obvious liability issues Baycol Vioxa both recalled because of rare side effects too rare to be detected in trials 0 B People in small communities exposed to some health hazard were not able to demonstrate the effect because there wasn t a large enough sample There are not enough sick people 3 Interactionscomplexity o Interaction means that behavior of 2 or more things together does NOT obey behaviors separately hard to do science when strong interactions exist because science works best but also when a factor has a consistent effect c Examples of interactions flash powder ammonia and bleach made together chlorine gas alcohol and driving is okay but you don t wand to mix them together alcohol and sedatives separate but don t mix together o Lots of interactions with diet and health part of the reason that advice on diet changes so often 4 Time lags slow progress o In a causeeffect relationship the effect occurs long after the cause It slows the rate at which we can cycle through the scientific method Familiar time lag adjusting temperatures Drug example DES drug given to suppress miscarriage 0 Used from the 1940 s1970 s caused reproductive problems and cancer in the children of mothers who took it 30 year time lag Hepatitis drug trial disaster phase 1 in 1993 o 5 patients died so many because there was a time lag between doing irreversible damage to patient lives and patient death 0 Time lags major problems in climate change research 0 Friday November 4 2011 Biological Determinism o Fundamental Question Nature vs Nurture Is it in our genes Are we genetically programmed to be who we are or can we control that 2 Themes humans make difficult subjects amp correlation does not imply causation Where relevant in today s society o Sexual preference Athletic abilityperformance Intelligence Addiction drug use Mental health Crime Weight amp eating habits Gender roles in society What difference does it make if behavior in genetics is culturallearned o Accountability o Prevention o Education o Rehabilitation History most notorious abuse of this issue was the Nazi regime in the 1930 s 1940 s o Individuals deemed inferior in some fashionnon productive were sterilized or killed youthinized o Nazis got their views on eugenics from the US showed us the sterilization flyer o In US forced sterilization for various mental deficiencies was continued into the 1970 s o Sterilization was done to prevent contaminating the gene pool 0 Assumes a genetic basis to the condition In the last decade or two a couple of multiplyconvicted rapists have been given freedom or reduced sentences in exchange for their gonads or some other biological modification treatment o In this case the assumed biological basis of raping behavior is neither genetic nor learned but chemical or hormonal How do we decide nature vs nurture o Nature biological determinism implies a genetic basis which means the behavior is inherited o Inheritance means that children resemble their parents 0 Smart parents have smart kids o Not easy to do in humans because 0 1 cannot do controlled breeding experiments 0 2 Parents strongly influence offspring environment Thus we usually have correlations to work with Sexual preference c We now recognize 2 dimensions to a persons sexuality from a behavioral perspective o A sexual preference who you want to have sex with o B sexual identity gender identity whether you think of yourself as male or female Estimated that 25 of men are gay and 12 of women are lesbian These proportions appear to hold crosscultures as best as one can tell Public attitudes about sexual preference 0 1970 survey in the US found that 43 thought that gay preference was learned Studies in humans of the biological basis of behavior have used 2 approaches 1 Look for anatomical and physiological coordinates of behavior 2 Look for a geneticinherited basis of the behavior Both have been guided by animal models Monday November 7 2011 Sexual preference Studies of inheritance o Rationale Identical twins share all genes full siblings share half of their genes people with adopted brothers share no genes 1 If sexual preference is entirely genetic then the rate of sexual preference similarity 100 for twins full siblings 50 and adopted 0 random So provides a suggestion of some inherited basis to sexual preference 2 Miscellaneous correlations weak correlations o Fraternal birth order probability that a man has homosexual preference increase with number of older brothers he has Cannot have genetic basis o Finger length ratio the ratio of the index finger to 4th finger is higher in women than in men In people with samesex preference there is a tendency for ratio to be lower than in heterosexuals of the same sex By this criterion samesex preference is associated with over masculinizing o OAE Otoacoustic emissions our ears make sounds too weak to hear Each ear has a characteristic frequency starting in early childhood Average frequency in men does not equal average frequency in women o Gay men have slightly OAE frequencies than heterosexual men Neuroanatomy various regions of the brain have been found to have small differences associated with sexual preference Maybe earliest study to find such a difference Simon LeVay 1991 Background work with rodents had shown that a brain region called the anterior hypothalamus controls sexual preference No difference between gay and heterosexual men for nuclei 124 but for nucleus 3 heterosexual male volume is greater than gay male volume which is about the same as female volume NEW SECTION Wednesday November 16 2011 Conflict is everywhere how is conflict manifested when it comes to uses of the scientific method When someone isn t representing the scientific truth of the situation Main topic Conflicting Goals o Scientific Method how it works to when it has problems and now to when not everyone wants it to work Covered a specific type of conflict called Tragedy of the Commons From now on we will address more obvious types of conflict A person s goals often depend on many factors 4 categories o A Material gain money power objects time fame o B Emotional factors 0 passion compassion we have strong feelings about relatives and loved ones and ex loved ones 0 Philosophy spirituality some religious groups oppose scientific teachings o Ego A person s reputation may influence their views we don t like to admit fault so we hold onto views we held in the past Examples Semmelweis 18181865 guy who worked in maternity clinic in Vienna and discovered that if physicians washed hands their hands before delivering babies they would prevent hundreds of lethal infections in the mothers His reports and book were not only rejected by the medical establishment but he was ridiculed by medical community because of ego o C Politics can get in the ways of seeking scientific truth 0 Lysenko and the suppression of Soviet genetics 1940 s1950 s Political ideology 0 There was a lack of constituency for certain projects AIDS as a disease of gays in the United States 0 The very funding of major science projects in the US driven by politics 0 watch video part II Prisoners of Silence Frontline 1993 Will observe various individuals and groups who do not want to give up FC n A parents for emotional reasons a B administrator Biklen who wants to build an FC institute not having it kills his dream a C hear of school admins who after having adopted FC will not admit its bogus Friday November 18 2011 Chapter 25 Talking about footprints of Bias generalities of bias Scientific method made to appear to give something other than the right answer We did conflict Manifestation of conflict in bias deliberate Someone does not want you to get the right answer In contest with supposed evidencebased evaluation How to recognize bias in someone s defense of an idea presentation etc o 1 Look for conflict of interest ex we don t expect tobacco products to say anything about the health effects of smoking o 2 Look for signatures of bias in either defenses of a study or challenges of a study long list in the book 0 a claim that all alternatives have not been disproved seen in the NOVA video horoscope experiment 0 b use anecdotes as evidence of generality or has representative the data not representative of the whole like in the video of FC where the parents would be like oh one time my child did such and such This proves for me that FC works FC part II video 0 c refusal to admit possibility that a model is wrong Bikklen did this we have to keep looking for the cases that are in favor of the model doesn t matter how many times it fails in general this attitude completely defies a nullmodel approach 0 d character assassination go after the person rather than their argument 0 e appeal to authority such and such holds this view you re incapable of holding your own view o f defend a model that cannot be shown wrong this is known as an unfalsifiable model true of many creationist arguments and in some cases the psychics different than c Note many of these biases are rampant in trials Ways to bias a study o 1 Before the study control the null model 0 ex safe until proven harmful vs harmful until proven safe makes a huge difference in a company o 2 In the design of the study 0 a assay for a narrow spectrum of results avoiding those that would be unfavorable to your goal ex might screen a drug or additive for cancercausing effects when you know it causes reproductive dysfunction ex FDA has for decades been careful to screen products against causing cancer but not screened products of behavior and reproduction b conceal true protocol studies are often blind on paper but not in reality or controls may be bogusflawed c Use small samples lowers the chance of observing rare events and thus makes certain statistical outcomes more likely o 3 During the study the datagathering phase 0 change design midcourse drug companies will initiate a study and then abandon if results look bad sometimes finish study and block production o 4 After the data are in bias the evaluation 0 a Search for statistical tests to support desired outcome 0 b throw out unwanted results very sinful common to throw out entire studies How to avoid bias o Publish protocols in advance prevents midcourse changes in the design Publish actual raw data and this allows someone else to do the analysis Specify evaluation criteria before you get the results less wiggle room consider having asked the FC supporters what they expected before the test Anticipate vested interest o See book 0 0 Monday November 21 2011 Medicine drugs and bias Many books written recently The Last Well Person Overdosed America Should I Be Tested for Cancer Overall strong biases in our medical establishment drug companies from the very diagnosis of conditions to treatments that has moved us toward over medication Drug company tactics and bias Sales who decides what prescription drugs are used c Your doctors or physicians How have drug companies influenced physicians 1 Provided incentives to prescribe company drugs free trips dinners outright cash 2 Persuasion and engendering personal obligations o Reps take physicians to lunch 0 Free dinners gifts 3 Expose physician to drug company logos and name to gain familiarity posters pencils notepads 4 Present biased information to physicians about product efficacy 0 Publish ads in medical journals to look like research articles 0 CME containing medical education courses required of physicians every so often are often sponsored by drug companies Bias in research specific example In study design Comparing unequal doses Astro Zoneca compared 40 mg of Nexium their new drug with 20 mg of Prilosec their old drug about to go off patent Nexium proved superior Inappropriate control pain killer Oxycotin used no drug as a control o Wrong age group drugs for diseases of aging often tested on younger people than will be the main clientele reduces side effects During the study o Pharmacia study of blood pressure meds was stopped 2 years early when it was realized that the drug was proving less effective than the cheaper alternative After the study o Hiring ghostwriters to write final draft actual researches authors may not be involved in writing at all Allows a spin on the results 11 of published studies Monday November 28 2011 Chapter 26 Brain Flaws Rational decisions not necessarily easy or feasible our brains unconsciously do the wrong thing in many circumstances 1 Optical illusions and magic tricks 2 Reinforcement we search for confirmatory evidence of what we already believe not the same as searching objectively for the truth 3 Draw causation from correlation done before 4 Respond to perceived risks rather than actual risks lotto ads take advantage of this by showing winners well out of proportion from the odds of winning o Also true that we respond to same information given different ways 1 in 10000 seems rarer than saying 10 people in greater Austin area 5 Respond emotionally to many factors fear greed body language appearance our control of a situation o urban legends spread because of emotional impact o witch crazes 6 Memory reconstruction our memories are rebuilt over time c We think we remember something clearly but are wrong eyewitness ID 7 Our preferences are strongly influenced by the context such as perceived relative benefit day 1 of class