Research Methods in Psychology
Research Methods in Psychology 031 010
Popular in Course
Popular in Psychlogy
This 115 page Class Notes was uploaded by Donny Block on Friday October 23, 2015. The Class Notes belongs to 031 010 at University of Iowa taught by Jonathan Mordkoff in Fall. Since its upload, it has received 22 views. For similar materials see /class/228096/031-010-university-of-iowa in Psychlogy at University of Iowa.
Reviews for Research Methods in Psychology
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 10/23/15
Swmyg p5y h gi FSEWU1 Outline three separate issues for surveys who is sampled how are they contacted what are they asked Convenience vs various kinds of Random Sampling Types ofSurveys Types of Items Surveys a structured set of items designed to measure attitudes beliefs values or behavioral tendencies items include direct questions agreedisagree statements fillintheblanks amp scales etc note the items do not have to concern the person taking the survey note also survey work not always respected Sampling general rule with regard to external validity you can only apply your results to the people who could have been sampled why didn t this issue come up before now most experimental work concerns the structure of the mind andor the laws that govern behavior these are assumed to be universal so the subjects employed can be anyone but surveys often concerns specific attitudes beliefs values etc which are personspecific Sampling Convenience Sampling use easilyrecruited subjects eg street Intro pool Probability Sampling when each person in the population has a definable probability of being sampled why this matters populations often have subpopulations groups within them eg divide pop by sex race etc to whom your results apply depends on who could have been sampled Sampling I two main versions of Probability Sampling 1 simple random sampling no attempt is made to ensure that the sizes ofthe groups in the sample match those in the pop just sample random people subtypes to get a 10 sample from 5000 people standard sample 500 people systematic sample every tenth person on list Bernoulli each person has 10 chance Sampling 2 stratified random sampling the sizes ofthe groups in the population are taken into account don t just sample random people subtypes proportional force the group sizes to match the sizes ofthe groups in the population nonproportional quota example force the group sizes to be equal to each other Sampling why use proportional stratified random sampling it s a lot like verifying successful random assignment it s only really important when samples are small it s done because by luck the sample might not match the population which would be a threat to external validity Sampling Why use nonproportional stratified random sampling it s only really important when some groups in the population are very small eg 5 or less it s done because some statistics require a minimum number ofobservations in every cell to be used it s also done because the standard error is reduced by 1N and it s best to have equal errors 10 Sampling why use nonproportional stratified random sampling target population the people you want to sample accessible population the people you can sample it s also sometimes done when stratified random sampling is used within an accessible population that doesn t match the target population this is not farfetched on some variables the Intro research pool doesn t represent the entire population very well 11 Sampling how do you choose a method ask yourself how important it is to have a sample that accurately represents the target population if not very convenience if sort of simple random sampling if very stratified random sampling 12 Sampling is there any way to make this easier if I said that matching the population was very important but the population is huge andor spread out yes cluster sampling when people are conveniently pregrouped via an irrelevant variable and random sets of these groups are used example want all undergrads use a set ofclasses want all Iowans use a set of counties 13 Sampling is there any way to make this more complicated proportional stratified weighted cluster sampling not ajoke actually used example equal number of rural suburban amp urban clusters proportional sampling on important variabes inside each cluster separately data from clusters reweighted for proportions of Iowans living in each environment 14 Survey and Item Types in one minute SurveyTypes facetoface phone written mailed electronic Item Types openended vs closed questions checklists Likert scales and ranged scales 15 SurveyTypes x Sampling one ofthe mostpopular ways to survey is by phone however some people Gen CP only have cells which are not included in randomnumber callers how do correct the data so you have an accurate estimate you separately get an estimate of how many people only have cell phones and also identify some attribute that is correlated with only having a cellphone use stratified random sampling via this attribute 16 Survey Types x Reactivity I the socalled Bradley Effect aka Wilder Effect is the difference in how people say they ll behave vote vs how they actually will behave vote in private it s an example of reactivity how can you avoid it you ask yourself how do researchers reduce this kind of reactivity in general by keeping the experimenter away from the subject therefore use automated datacollection for this situation 6 M5 Mam635 SE vn4mu m m 1 52f w Ef gy LEWUEEWQQ Predictions of Theories revisited I if X then Y theory after transation X causal conditions are met therefore Y so predicted data should be found often Y is a single event under specific conditions i please bisect this line Predictions of Theories revisited I if X then Y theory after transation X causal conditions are met therefore Y so predicted data should be found Y can also be a pattern ofdata across conditions as the number of inactive bystanders is increased the odds andor the speed of a person the subject helping someone should decrease Predictions of Theories revisited if X then Y X therefore Y if X is varied then Y should vary X is varied therefore Y should vary in a certain way Outline two classes of variables three ways that variables are treated I various definitions of experiment Classes ofVariables 1 Manipulated things that are completely under the control of the experimenter and are set by the experimenter things that do not depend in any way on the subject eg lighting conditions task difficulty instructions 2 Measured everything else things that are at least partly determined by or builtin to the subject things that cannot be known in advance eg attitudes response time gender helping behavior Types of Manipulated Variables 1 Situational features of the environment egl ofwitnessesl lighting background noise 2 Task elements of What subjects are asked to do egl easy vs hard trials 3 Instructional elements of how subjects are asked to do the task egl use imagery vs rote memory Note these distinctions have no effect on design or analysis Types of Measured Variables These two types are fuzzy any particular measured variable is moreorless of one type 1 stable builtin chronic permanent eg gender handedness major depression difficult to impossible to manipulate ethically these are often referred to as subject variables 2 labile situational acute temporary eg response time attitudes mood relatively easy to manipulate ethically the data are always this type Ways that Variables are Treated 1 Potential Cause thing of interest that could directly or indirectly help to determine the value of another variable of interest eg lighting conditions task difficulty instructions also attitudes gender 2 Effect measured thing of interest that could be influenced by the potential cause eg response time helping behavior attitudes Ways that Variables are treated 1 Potential Cause thing of interest that could directly or indirectly help to determine the value of another variable of interest 2 Effect measured thing of interest that could be influenced by the potential cause 3 Extraneous potential cause that is not of interest note effects that are not of interest have no name and are ignored Extraneous Variables Given that extraneous variables by definition are things that we re not interested in why do we care about them Because they are potential alternative causes of any effects that are found I threat to InternaIValidity Thursday Extraneous Variables eg study of diffusion of responsibility effect slowerlesslikely helping behavior potential cause of interest of witnesses extraneous diversity of witnesses when you increase the of witnesses you almost always also increase the diversity ofthe witnesses maybe the real cause of slower helping is witness diversity instead ofwitness numerosity Extraneous Variables eg study of diffusion of responsibility effect slowerlesslikely helping behavior potential cause of interest of witnesses extraneous diversity of witnesses This is an example of confounding A confound noun is an extraneous variable that covaries with the potential cause of interest Without further research you cannot know for sure whether the Potential Cause or the confound is the actual cause ofthe effect Confounds What should be done I 1 control all extraneous variables don t let them vary at all obey the all else being equal rule I 2 measure and analyze those that you can t control remove their influence statistically by switching to analysis of covariance will not be covered in this course One last point on Ways that Variables are treated The exact same variable can be a Potential Cause an Effect or Extraneous in different experiments eg affective positive or negative mood a mood gt reca performance b reca performance gt mood c distraction I mood gt reca performance The Prototypical Experiment Has one measured variable that is treated as the effect of interest referred to as the dependent variable DV Has at least one manipulated variable that is treated as the potential cause of interest referred to as the independent variable IV levels ofthe IV create the conditions Definitions of Experiment 1 Strict must match the prototypical experiment ie must have at least one IV that is manipulated 2 Middle ground strict experiments are called true experiments those with subject vars are quasiexperiments 3 Loose manipulated IVs may be replaced by subject variables but not by labilesituational measured variables Who ca res about the definition of Experimentquot Thursday I the key is InternaIValidity for now a preview using pictures Mostbasic Experiment potential cause effect IV gt DV manipulated measured Basic Controlled Experiment potential cause effect IV gt DV manipulated measured pessible eonFounds EVs held constant Basic QuasiExperiment potential cause effect SV gt DV measured meaSUFEd Basic QuasiExperiment main problem potential cause effect measured meaSUFEd possible confounds EVs highly numerous and hard to hold constant 45W pgydmwgiga r MAM Outline P X E Designs revisited Interrupted Time Series Designs P x E Designs revisited Always involve at least one factor that is not manipulated ie the person factor eg race ofsubject x sameldifferentrace roommate handedness X stimulusresponse compatibility anything involving sex differences P x E Designs revisited Two ways to run these quasiexperiments 1 take one sample and split the data after the fact in terms ofthe person factor label eXpostfacto quasiexperiment 2 take two samples one inside each level ofthe person factor label planned quasiexperiment P x E Designs revisited Why does the distinction between eXpostfacto and planned quasiexperiments matter answer depends on how many person factors are going to be included Ifthere is only one person factor then the only important difference is planned quasieXpts have equalsized groups eXpostfacto quasieXpts often do not therefore planned usually have more power P x E Designs revisited Why does the distinction between eXpostfacto and planned quasiexperiments matter answer depends on how many person factors are going to be included Ifthere is more than one person factor then planned quasiexpts can be analyzed using ANOVA because equalsized groups implies no relationship between levels of different person factors eXpostfacto quasiexpts cannot because separate person factors are rarely unrelated P x E Designs revisited Example with two person factors manipulated IV stimulusresponse compatibility PF1 handedness note lefthanders often show smaller SRC effects PF2 athleticism note athletes often show smaller SRC effects Problem for the eXpostfacto approach lefthanders are more likely to be athletic there s a correlation between handedness and athleticism P x E Designs revisited Example with two person factors PF1 handedness PF2 athleticism table of N5 out of 200 people sampled athletes nonathlete righthanded 50 130 lefthanded 10 10 P x E Designs revisited I Summary ifyou only have one person factor then you can use either approach planned or expostfacto planned takes more work but planned has more power for a given total N ifyou have two or more person factors then you must used a planned quasiexperimental design or you have to switch from ANOVA to hierarchical regression 10 Interrupted Time Series Design I Many realworld interventions involve applying some treatment to a group of people and looking for a change in their behavior egl try a new way of teaching subtraction introduce a new antibingedrinking program The problem is the automatic confound between beforevsafter treatment and timeoftesting One solution to this problem is to use a non equivalent control group in some other location Another solution is to use an interrupted time series 11 Interrupted Time Series Design whatever you re interested in is the drop in the DV due to the treatment or due to anything related to time N i time treatmentTapplied Interrupted Time Series Design whatever you re interested in treatmentTapplied time 12 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 13 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 14 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 15 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 16 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 17 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 18 Interrupted Time Series Design whatever you re interested in treatmentTapplied time 19 20 Interrupted Time Series Design Ifyou have access to a second group with little or no possible crosstalk between groups add a nonequivalent control group apply the treatment to both groups but at different times staggered time series Interrupted Time Series Design whatever you re interested in treatment T applied T treatment applied time 21 22 Interrupted Time Series Design fyou have access to a second group with little or no possible crosstalk between groups add a nonequivalent control group apply the treatment to both groups but at different times staggered time series fyou don t have access to second group of subjects you can add additional measures that should not be affected by the treatment to act as a kind ofwithin subjects control condition something you re not interested in whatever you re interested in treatmentTapplied time Interrupted Time Series Design 23 mwwm m m ggg g S9 U E ENEmEEam True Score Theory an observed value X is the linear combination of the actual true value T and various errors e the expected value ofe is zero the value ofany particular e is not related to T assuming that T is constant the variability oins therefore equal to the variability of e thus the unreliability of a measure is due to e True Score Theory first equation let X score on some measure N the number of scores Xs that you have Y the mean ofthe N scores varX varX N unreliability of unreliability ofX N Apples and Apples theory of interest operational definition actual raw simple measures 3 coherent of subjects pre39process39ng set of data Outline Descriptive vs Inferential Statistics Summarizing Data Descriptive vs Inferential Stats Descriptive Stats summarize a given set of data the set of data is usually a sample not the entire population because these are just summaries they can t be wrong Inferential Stats go beyond a given set of data and make a probabilistic statement about the population from which the sample was taken the statement is a best guess about the population because it s a guess of some sort it can be wrong Summarizing Data Function to pass lots of information in few words Center Spread and Shape of Distribution Center mean median or mode Spread standard deviation variance IQR or range Shape skew amp kurtosis or name Summarizing Data Center arithmetic mean 121 1132 I JEN N Spread standard deviation PIquot 1 i V g I 2 a N 1 tr W I 1 1 1 Summarizing Data Shape name pm Jmshaped Hanna Rectangular lmfk almndal Pusii ve v ritghtj skew Negative left skew Summarizing Data A Mama why are Normal distributions special ifthe shape is Normal then estimates ofthe mean and standard deviation are independent this allows us to focus on one thing at a time most inferential statistics assume that the shape of all critical hypothetical distributions are Normal this assumption enjoys empirical support under certain conditions it must be true Summarizing Data A Example average of N dice rolls Rectangular N1 N2 N10 as N increases the shape ofthe distribution of means becomes Normal Summarizing Data xi Why do people like samples of 30 or more J shaped N 1 N if N is at least 30 the shape ofthe distribution of means must be Normal theory of interest operational definition actual raw simple measures 1 coherent of subjects pre39process39ng set of data Exam Need 2 pencils for multiplechoice part Might also want a pen for shortanswer part Please disable phones before 230 or don t bring Please have your student ID with you Studying hierarchical chunking amp dualcoding find or make pictures that organize large amounts ofthe information 1 make sure you can add in all ofthe relevant sub components subissues 2 make sure that you can reproduce all technical definitions 3 make sure that you know how some things depend on other things theory of interest actual raw simple measures coherent set ofdata of subjec psychology is an empirical science the theory of interest determines the useful data theories make predictions via operational definitions operational definitions are moreorIess valid Construct Validity Content ex aUStNe 31533 Predictive and face correlations Criterion Validity Discriminant Validity eXCIUSIve Validity why might these fail selfesteem selfconfidence theory of interest actual raw J simple measures coherent of subjects set of data we usually have to condense or summarize the raw data condensation also gt the multifaceted nature of constructs summarizing provides a link to True Score Theory Multiplechoice Questions treat each option separately I ifthere are all of the abovequot andor none of the abovequot options be logical if one option is definitely wrong then all of the abovequot can t be correct if one option is definitely rightl then none of the abovequot can t be correct Multiplechoice Questions Psychologists are empirical scientists therefore A they are dedicated to creating an empire based on science B they use data to test whether their theories make the correct predictions C they use rational logic to prove that their theories are correct D all of the above A measure A is either valid or invalid B is valid to some degree that can vary from zero to perfect C only has validity with regard to measuring some theoretical construct D can only be valid if it s not unreliable some more definitions Experiment strict must have at least one manipulated variable IV Correlational Study all of the variables are measured one is treated as the predicted variable the others are treated as the predictor variables the difference is important for two reasons different methods ofanalysis different threats to interpretation some more definitions I Middle ground between Experiment and Correl Study the P x E design analyzed like a strict experiment via ANOVA but P X E designs are limited to one SV any situation involving more than one measured variable as potential causes or predictors cannot be properly analyzed or interpreted in the same way as an experiment Outline Correlations that s it for today Correlations can be calculated between any two variables when both variables are naturally quantitative iel both are interval or ratio scales no orders egl height and respectability the standard correlation coefficient is used the same goes when one ofthe variables is dichotomous iel can be coded as Os and 15 eg height and gender but now it s called a pointbiserial correlation Correlations can be calculated between any two variables when both variables are naturally quantitative ie both are interval or ratio scales no orders eg height and respectability the standard correlation coefficient is used and the same goes when both ofthe variables are dichotomous eg gender and asking for directions tee hee but now it s called a phi coefficient Correlations can be calculated between any two variables when at least one ofthe variables is qualitative and takes on more than two values eg race religion or hair color then a different procedure must be used called multiple regression not covered in this course but it s still a type of correlation Correlations can be calculated between any two variables when at least one ofthe variables is ordinal eg order of arrival or birthorder position then a nonparametric procedure must be used which are also not covered in this course but they re still correlations Correlations look at regular correlations using a scatterplot Estimated IQ 130 120 110 100 6o 64 68 Height in inches 72 76 since the best fitting circle has a positive slope this is a positive correlation Correlations 10 look at regular correlations using a scatterplot Estimated IQ 130 120 110 100 6o 64 68 Height in inches 72 76 these data fit inside a tighter oval so this correlation is stronger 11 Correlations look at regular correlations using a scatterplot 130 7 120 7 0 these data form 9 a nearlystraight 390 110 7 line a O 45 o sothls correlation E 100 7 o is nearly perfect quota z 100 LIJ 90 80 i 56 6o 64 68 72 76 Height in inches 12 Correlations look at regular correlations using a scatterplot 130 7 120 Here s a moderate 9 negative correlation 390 a 110 4 U E 100 t 3943 U LIJ 90 80 16 20 24 28 32 Body Mass Index 13 Correlations only linear relationships are measured 13o Strength 00 O these data show a very clear pattern but the correlation is actually zero 10 20 3o 40 50 6o 70 Age 14 Correlations can be calculated between any two variables provide a measure ofthe linear relationship only symbol r varies between 100 and 100 also provide a measure of how much ofthe variance in one variable is explained by the other variable symbol r2 name coefficient ofdetermination varies between 000 and 100 are greatly affected by the range of values 15 Correlations are greatly affected by the range of values 130 H N O H H O 100 Estimated IQ k0 O 80 Fullrange 070 Restricted range 000 5 6 and 5 7 only 6o 64 68 72 76 Height in inches 16 Correlations are greatly affected by the range of values Estimated Strength 130 120 110 100 o Fullrange000 Restricted 39 range 100 39 minors only 10 20 30 40 5o 60 70 Estimated Age 17 Correlations can be calculated between any two variables provide a measure ofthe linear relationship only symbol r varies between 100 and 100 I also provide a measure of how much ofthe variance in one variable is explained by the other variable symbol r2 name coefficient ofdetermination varies between 000 and 100 are greatly affected by the range of values cannot generalize outside ofthe measured range w im i p5y h gi mh Setup What s the difference between a large and smallN design nope that s too obvious try again answer whether you look at averaged data and analyze all subjects at once vs you look at and analyze each subject separately Outline Why use a smallN design Main types ofsmaIIN designs Main problems with smallN designs Reasons to Use a SmallN Design 1 Summaries can be misleading individualsubject validity the extent to which averaged data match those from individuals main threat smearing Reasons to Use a SmallN Design smearing 00 100 Percentage correct Percentage correct 50 L 50 Trials gt Trials gt FIGURE 115 Concept learningr outcomes as predicted by a continuity and b noncontinuity theory Reasons to Use a SmallN Design smearing Percentage correct OO Trials gt FIGURE 116 How grouping data from individual children in a concept learning experiment can produce a smooth but deceptive learning curve Reasons to Use a SmallN Design 1 Summaries can be misleading individualsubject validity the extent to which averaged data match those from individuals main threat smearing In general interest in a nonlinear relationship with different constants for different subjects 2 Potential subjects are rare or difficult to recruit 3 Detailed understanding ofa few sometimes preferred over coarse understanding of many Types of SmallN Designs SmallN designs can t have control groups so all subjects must act as their own control so all smallN designs use repeated measures often backandforth more than once Basic examples 1 Withdrawal aka reversal designs baseHne treatment baseline again the above is an ABA design Types of SmallN Designs 1 Withdrawal aka reversal designs baseHne treatment baseline again treatment again the above is an ABAB design 10 Types of SmallN Designs 2a multiple baseline aka staggered designs IV 1 IV 2 IV 3 baseline baseline baseline treatment baseline baseline treatment treatment baseline treatment treatment treatment note the treatments can either share a DV or have separate DVs especially useful design when withdrawal would be unethical andor ineffective Types of SmallN Designs 2b multiple baseline across subjects designs this is the same as a staggered time series quasi design with only one subject per group 11 12 Types of SmallN Designs 3 changing criterion designs in general any design Where the treatment is adapted to the subject example no cold turkey target number ofX where X is some bad thing during a given phase of the treatment depends on numberrate ofX during previous phase start with Phase 2 ofthe patch if lt loday Types of SmallN Designs 4 various combinations example multiple baselines withdrawal multiple changing criteria 13 14 Problems with SmallN Designs 1 No builtin reason to assume generality 2 No formal way to deal with betweensubject differences statistically 3 No formal way to deal with betweensubject differences theoretically 4 Impossible to conduct an exact replication 5 Often fail to include conditions needed to test for interactions