Research Methods PDF

Two primary types of research: (but both ultimately have the goal to either describe, predict or explain a phenomenon) ○ A) Basic research Main goal: to understand psychological processes Knowledge doesn't have to be applicable, the importance is...

Two primary types of research: (but both ultimately have the goal to either describe, predict or explain a phenomenon) ○ A) Basic research Main goal: to understand psychological processes Knowledge doesn't have to be applicable, the importance is in GETTING it in the first place Not really focused on solving some kind of a problem ○ B) applied research Finding solutions to problems Usually uses knowledge acquired by basic research and provides new questions for it Understanding and solving a problem of immediate concern Sometimes leads into EVAULATION research Aka program evaluation Behavioral research methods used to study effects of certain programs on behavior E.g. How is the new educational program doing? Is it effective? If it works/doesn't work, why? Predicting behavior ○ E.g. An employees' job performance based on tests and interviews, which variables predict dangerous criminals,… Explaining behavior ○ It's cool and all that we can predict someone is gonna be dangerous, but WHY The researchers greatest enemy - common sense ○ Can be blinding and prevent the scientists from seeing alternatives Why is studying research methods useful? ○ Applicable to any field, all professions need to do some kind of research to be succesful in their carriers ○ Psychology is a science ○ Helps to make everyday and big decisions ○ Development of critical thinking ○ Helps one become an authority - have great knoweledge about a certain topic and become superior WHAT MAKES AN INVESTIGATION SCIENTIFIC? Aka the scientific approach 1. Systematic empiricism 1. Empiricism - relaying on observations to draw conclusions 2. Observation done in a way o draw valid conclusions, not just 'in general' or in a casual way -> it's done in a 'systematic' way 3. E.g. Carefully crafting specific studies to obtain data 2. Public verification  1. The methods and results must be accessible to others, so the research can be observed, duplicated, verified 2. Making sure the phenomenon is real and not just some kind of fabrication 3. Helps to find possible errors and how to fix them 3. Solvable problems 1. Only investigate the questions that are answerable at the current time Pseudoscience does not follow these criteria: ○ Data is not collected in a systematic fashion ○ Based on personal experiences and myths, biased data ○ Not possible for verification ○ Theories that cannot actually be tested THE TWO JOBS OF A SCIENTIST Discovering, describing, and documenting new phenomena, patterns and relationships before testing out hypotheses = conducting research in the CONTEXT OF DISCOVERY Developing and evaluating explainations of the phenomena ○ Theories to explain patterns, testing those theories ○ Theory = a set of propositions that attempts to explain the relationship among a set of concepts, WHY and HOW are things connected Has to be supported by empirical data Proposes a causal relationship Is coherent and clear, consistent, logical Uses as few concepts as possible Generates testable hypotheses Stimulates other researchers Solves an existing theoretical question ○ Model = HOW are things connected, not WHY Doesn't try to explain but rather just describe ○ Any explaination that is made after the fact is not satisfactory (=post-hoc expl.) It always easier to try to find things that support our claim after we have proved the theory to be true (finding signs that it was truthful before, etc.) Basically, things should be always predicted BEFORE actually doing the experiment which should only say if it's true or not To provide a convincing test of a theory - hypotheses a priori (aka before conducting the experiment) ○ Theories themselves cannot be tested, the propositions are too broad and/or complex to be tested directly -> that's why we test hypotheses Deduction - derived hypotheses (specific implications) from a theory (a broad proposition) If theory true, what will we see/observe?  Induction - hypotheses from a number of facts, previously seen patterns, etc. Called "empiric generalizations" Usually we don't even have a theory why certain variables are connected, we just know that they might be and are trying to prove it All hypotheses have to be open to falsification (there has to be a way we could potentially find out if they are incorrect) ○ The terms have to be clearly defined Conceptual definitions - like in a dictionary, not the best for research Operational definitions - specifically stating how the concept is measured in a particular study (therefore, other researchers can then used the same 'measurement' when reevaluating that experiment) Scientists try to find many ways to prove theories and find strong data that supports them - usage of METHODOLOGICAL PLURALISM ○ Aka using a lot of different methods Theories are usually formed in such a way that they contradict other already existing theories - so we have a clearer answer ○ Strategy of strong inference Why can't we technically prove or disprove a theory? ○ Logical impossibility of proof - obtaining empirical support for hypotheses does not mean the theory was correct in the first place  Hypotheses can be true even without the original theory being so as well ○ Practical impossibility of disproof - hypotheses can be disconfirmed by the data, but the theory might be true There are many factors that can effect the result of our testing (like incorrect data collecting, insufficient sample size, etc.) Journals refuse to publish so called 'null findings' Aka variables we found out are not related to behavior because are they really? Or do we just suck at our job? DIFFERENT TYPES OF RESEARCH: 1. Descriptive research ○ Describing behavior, feelings, etc. Of a particular group of individuals ○ Animals in their natural habitat, public opinion polls,.. ○ No desire to relate the behavior to certain variables ○ Provides the foundation for other research 1. Correlational research ○ Examines the correlates or causes of behavior ○ Is there a certain relationship between two variables? And if yes, what kind? ○ But doesn't concern itself with whether a certain variable causes the other 1. Experimental research ○ Do certain variables cause some kind of a change? ○ Researchers changes one variable (independent variable) and sees if it behavior (dependent variable) somehow changes 2. Quasi-experimental research ○ In cause and effect relationships researchers wish to control all other factors that might influence the results, but its not always possible ○ This research method allows for the event to occur naturally or manipulates an independent variable but doesn't concern itself with the other factors ○ Conclusion much less reliable, but still useful 3. (Animal research) ○ Studying similarities and differences in human and nonhuman animals functionality ○ Can be raised in laboratory conditions -> eliminates a lot of biasing factors ○ Studied under an extensive time and controlled conditions ○ E.g. Basic motivational system, pain and pain relief, evolutionary perspective on mind and intelligence, substance abuse,… ○ Ethical issues THE EMPIRICAL CYCLE Shows how science develops over time 1. Observation ○ An idea for a research question 2. Induction ○ Formulating a specific theory based on observable facts ○ Coming up with a THEORY ○ "leap of faith" 3. Deduction ○ = logic ○ Predicting following from theory, forming hypotheses ○ Working with DEFINITIONS 4. Testing ○ Actually collecting data ○ Analyzing data and making conclusions 5. Evaluation ○ What does the result tell me about my theory ? ○ Adjusting, improving? ○ Critical review - was there anything I could have done differently? BEHAVIORAL VARIABILITY AND RESEARCH Schema = cognitive generalization that organizes and guides the processing of information ○ We all make generalizations about certain topic that are unique to us and can be very different from the ones made by others ○ Our reactions and decisions are greatly affected by our schemas (e.g. Who am I going to vote for president? My schema for a president is someone wise, well educated,…) ○ Helps us organize and process information - increase our efficienty and effectivness Variability ○ = describing, predicting and explaining differences in behavior and mental processes between people ○ All aspects of research processes revolve around it ○ Usually what psychologists and behavioral researchers actually study (aka how and why behavior varies between people, stimuli, over time, etc.) People act differently in certain situations (sunny vs. Gloomy day), react differently to the same situation (extrovert vs. Introvert), behave differently over time (dating is gross to a kid vs. Desirable for an adolescent) ○ All research questions are about behavioral variability  Research should be designed in a way that allows the researcher to ○ answer questions about beh. Var. ○ Any kind of measured number assigned to a behavior needs to be connected the behavior itself in a meaningful way Aka the scores and data we collect have to actually match the reality of variability in the behavior ○ All data collected needs to be studied - we have numbers that vary and we need to understand why and somehow make sense of them -> for that we use STATISTICS Descriptive statistics - summarizing, transforming large number of scores into averages/percentages,… Inferential statistics - used to determine the generalizability of findings Questions like how likely this is a coincidence or if I had other factors affecting the research or how representative are my findings to a larger population? Terms connected to variability ○ How much variability is there in a set of data? How much variability is there between the scores and the mean -> variance ○ The difference between the largest and the smallest score -> range ○ The average around which the numbers accumulate -> the mean The total variance = systematic + error variance ○ Important to figure out what variables are related to each one ○ Systematic variance The part of total variability in behavior that is actually related in a predictable fashion to the variables investigating Aka the stuff that's actually related and is not a difference between results based on chance Seen when behavior changes in a systematic way that somehow corresponds to changing the variables ○ Error variance Shows how other factors (that the researcher is not studying) affect the data If it's too large - can make it difficult to actually distinguish if the variables of interest are actually related to variables in behavior ○ How do we tell them apart? Statistical analyses Effect size ○ Shows how strongly things are related  ○ Done by comparing the proportion of systematic variance to the total variance ○ Can be anywhere from 0.0 (zero correlation) to 1.0 (perfect relationship) Meta-analysis ○ Research can never be perfect individually - we achieve some kind of result that is only true for our specific study (if smn else did the exact same thing, they might get different results) Because of a large number of other variables affecting our behavior (mood, age, how we speak to the participants,…) ○ -> meta-analysis - used to analyze and combine results from a large amount of individual studies ○ Statistically combining effect sizes ○ Helps us not only see how strongly things are actually related but also to examine what other variables might influence our behavior THE MEASURMENT OF BEHAVIOR Goal of obseration = putting people's thoughts behavior/attitudes into categories Measuring = assigning numerical values to the categories, whih you use in calculatiions Collection of values has to be: ○ Exhaustive - minimally one value for each observation ○ Mutually exclusive = maximally one value for each observation (no option for more than one answer to apply to us) Three types of measures: ○ Observational measures Direct observation of behavior - anything the participant does that can be observed ○ Psychological measures - relationship between bodily processes and behavior Specialized equipment to measure reactions not observable by the naked eye ○ Self report Replies to interviews and questionnaires Info about one's thoughts, feelings Cognitive self reports - what people think of smt Affective self reports - how they feel Behavioral self reports - how they act Psychometrics - a specialty devoted to study psychological measurement Scales of measurement  ○ Responses are always turned into numbers and analyzed = how numbers used to represent participants' responses correspond to the real number system Nominal scale - numbers as labels ○ They indicate attributes (aka if you are married - number 1, if not - number 2 etc.) ○ No sense to perform mathematical operations on them Ordinal scale - rank ordering ○ Ordering the participants into some list, but hard to distinguish the distance between e.g. 1st place and 2nd place Interval scale - differences between numbers correlate to the difference between participants in the characteristic ○ E.g. IQ test, score 90 and 100 are in the same relation to one another as a score 110 and 120 ○ However - no true zero point, the score of 0 doesn't mean the characteristic is not present (e.g. Fahrenheit scale) ○ Numbers cannot be multiplied or divided Ratio scale ○ Have true zero point, can perform all mathematical operations on them ○ E.g. Weight, price Determine the amount of information provided Help distinguish what statistical analyses can be done Reliability of a measure ○ We want the variability of the numbers to reflect the variability in characteristic/response measured ○ Reliability - the consistency/dependability of a measuring technique Measurement error - anything that distort the observed score (mood, malfunction,…) Observed score = true score + measurement error True score - if the measure was perfect Types: Affected by transient states of participants - health, fatigue, mood,…  Stable attributes - less intelligent participants may misunderstands, paranoid people may alter their responses, different motivation Situational factors - room temperature, researcher's attitude,… Characteristics of the measure itself - too long of a study, too ambiguous questions,… Mistakes in recording Combining the sets of scores of many participants and calculate the variance, we find out its made out of the variance due to true scores (systematic variance, how results differ between each other) and variance due to measurement error (error variance) -> to asses reliability - estimating the proportion of total variance that is systematic vs. Error Reliability = true score variance/total variance Ranges from.00 to 1.00 Types of reliability Test-retest Many retakes of the same measurement that should be constant over time But how stable is it over time? Parralel Uses similar measures with the same participants But are the measurement instruments parralel enough? Interitem For measures consisting of more than one item Asses the degree of consistency among items on a scale (aka multiple questions are summoned into a score that's then seen if it correlates to the aspect of behavior) E.g. When measuring extraversion, we want all responses to show some aspect of extraversion Looking for total correlation for each question and the final score and comparing each correlation to each other if they are on the same level Index of interitem reliability - split-half reliability Separating the scores into two groups and comparing the total score of each set Cronbach's alpha coefficient - the equivalent to the average of all possible split half reliabilities Interrater  Consistency among two or more separate researchers Increasing the reliability Standardize administration of the measure - same conditions for everything Clarify instructions and questions Train observers Minimize errors in coding data Most estimates - done by examining the correlation between two measures that are suppose to be of the same behavior -> correlation coefficient (positive - direct relationship, negative - indirect relationship) Validity of a measure ○ = extent to which a measurement procedure actually measures what it intended to measure ○ Do the scores on the measure relate to the behavior ○ Something can be reliable - many studies coming to the same results - but doesn't have to be valid in the sense we are measuring what we think we are ○ Types Face validity Involves the judgement of the researcher/participants Aka does the thing I do make sense logically? Construct validity Entities that are inferred on the basis of empirical evidence (even if they cannot be directly observed) Things like intelligence, attraction, moral maturity,… Using construct validity to see how a certain hypothetical construct relates as it should to other measures Measure should contain convergent validity (correlates with measures that it SHOULD correlate with) and discriminant validity (not correlate with measures it should NOT correlate with) Criterion-related validity Distinguishing among participants on the basis of a particular behavioral criterion Examining behavioral outcomes that the measure should be related to Concurrent validity - two measures administrated at the same time, scores of measures related to behaviors as they should RIGHT NOW  Predictive validity - longer time period ○ Statistical validity - was data analysis done correctly? ○ Internal validity - are alternative explanations rules out? ○ External validity - is the result generalization? Fairness and bias ○ Test bias - when measure id not equally valid for every test taker APPROACHES TO PSYCHOLOGICAL MEASUREMENTS Observational approach ○ Direct observation ○ Questions like: Will the observation occur in a natural or contrived setting? Naturalistic - behavior occurs normally with no intrusion or intervention Settings now specifically set up, daily activities E.g. Participant observation - researcher partakes in the same activities as the subjects Negatives - becoming biased towards the group, group senses you are weird and behaves differently Contrived - in a setting specifically arranged for observing and recording (observers are usually behind a one way glass or behaviors are being videotaped - lab experiment + situations set up artificially in the real world - field experiment) Will the participants know they are being observed? Undisguised observation vs. Disguised observation Reactivity - the fact that people behave differently when they know they are being observed Ethical issues - consent, privacy,… -> partial concealment strategy - letting participants know they are being observed but not saying how or why -> Recruiting knowledgeable informants (people close to subjects) - to observe and rate their behavior -> Unobtrusive measures - taking measurements that can be obtained without participants knowledge and without any intrusion E.g. Which parts of the book find students interesting? -> look at the highlighted parts How will participants' behavior be recorded? Narratives - full description of one's behavior, unstructured observation method  Usually taken in field notes because full record of every behavior is almost impossible But it has to be content analyzed so it can be statistically studied Checklists - structured observational method Recording attributes of participants and whether certain conditions were met Using operational definitions Temporal measures When a certain behavior occurred and for how long How much time has passed between an event and a behavior = latency Reaction time - mainly used by cognitive psychologists - how long it takes for info to process in the brain Task completion time - how long does it take for participants to solve a problem Interbehavior latency - time elapsed between the performance of two behaviors How long behavior lasted = duration Observational rating scales Measuring the quality or intensity of behavior Usually involves using a scale with unambiguous criteria ○ Increasing the reliability Interrater reliability = the degree to which the observations of two of more independent raters (researchers) or observers agree Providing clear and precise operational definitions Using the coding system and discussing their practice ratings with others Physiological and neuroscience approaches ○ How processes in the brain correlate with psychological phenomenon ○ Types: Measures of neural electrical activity EEG, EMG Neuroimaging Structural - examine the physical structure of the brain Functional - fMRI, identifying areas where psychological functions occur Measures of autonomic nervous system activity Measures of heart rate, respiration, blood pressure, temperature,…  Blood and saliva assays Certain hormones and their effect on behavior Relationship between psychological processes and health/illnesses Precise measurement of overt reactions Special equipment like special sensors, blood flow to certain areas Self-report approaches ○ Questionnaires and interviews ○ More than questions, asking participants to rate statements about their attitudes or characteristics ○ Instructing to make a list, rate how they feel, describe their feelings ○ Any prompt that leads to some kind of an answer = item Items designed to be analyzed by themselves or as a part of a multi-item scale Single-item measures - easy questions that do not require any other supporting ones (e.g. What is your age?) Multi-item scale - more items to achieve a more reliable response (e.g. How are you satisfied at school? + how at home? + how with your love life? Etc.: how is a person overall satisfied with their life) How to write them so misinterpretations don't happen? Be specific and precise in phrasing Write items as simply as possible - avoid complex words and phrases Avoid making unwarranted assumptions (usually based on our own experiences) Conditional info should precieve the key idea E.g. Use: "If a good friend were depressed for a long time, would you suggest he or she see a therapist?” and not: “Would you suggest a good friend see a therapist if he or she were depressed for a long time?” (so people don't start to formulate an answer before the end of the question) Don't use double-barreled questions Ask two questions but offers only one response Appropriate response format Free-response format - provides an unstructured response Problems: participant doesn't know what response we are looking for, how extensive, we have to use content-analyzation Rating scale response format  Have to correctly choose the labels/numbers /the scale? Easy to use Multiple choice/fixed-alternative response format Influenced by the answers provided (just as in rating scales) Special type is true-false response format Pretest the items See how the exam behaves before offering it to participants ○ When using questionnaires - we can use information about past measures or rework already existing ones Journal articles Books about measures - more broad or in our specific field Data-bases available on the internet Some questionnaires may be purchased ○ Interviews Contains an interview schedule - series of items used during it Consider how the process of the interview might affect the responses How to improve quality: Friendly atmosphere Build trust and likeness Maintain an attitude of friendly interest Appear interested in the responses Conceal personal reactions Never show any emotion regarding the answers Order the sections of the interview to facilitate building rapport and logical sequence Start with basic topics and work your way up Ask questions exactly as worded Same way to every participant Do not lead the respondent Not putting your words in their mouth ○ Questionnaires vs. Interviews Questionnaires do not require much training, less expensive (assessing all the people at once) + anonymity is assured -> more honest answers But inappropriate to illiterate people, young children, severely disturbed individuals, etc. In interviews researchers can be sure the respondent understands each item + elaborate details  ○ Biases The social desirability response bias Participants concerned with how they will "look" -> response in a socially desireable way rather than truthfully Solution - word items as neutrally as possible, assure that resposes are anonymous, observers should remain unobtrusive Acquiescence and nay-saying response bias Tendency to agree with statements regardless of content (acquiescence) and vise versa (nay-saying) Using the reverse polarity - using two questions that are opposite of each other Archival data ○ Using research data collected prior to the time of conducting the research ○ Studying social and behavioral changes over time, insight on how people acted/felt in the past ○ Some archival is also necessary for certain research topics studied even today ○ For studying events that have already occurred which we cannot conduct and prepare atm ○ Some researches need a large number of data ○ But concerns with reliability and validity Content analysis ○ Converting written/spoken material into meaningful data that can be analyzed ○ = set of procedures that classifies units of text into categories relevant for research ○ First, finding the utterance (theme) ○ Then, decide if I'm gonna classify units into exclusive categories or rate them on some specific dimension ○ Being very specific with the classification, testing the interrater reliability SELECTING RESEARCH PARTICIPANTS Researchers have to pick a sample of the population ○ Process called sampling Behavioral researchers actually rarely ever use random samples Discrete variables - také only a limited number of values (e.g. Gender) and continuous variables - any value on the scale (e.g. Age) Distinguishing between quantitative data (results from any sort of measurement) and categorical/qualitative data (categorizing things and data consists of frequencies for each category)  probability samples - the likelihood that any individual in the population will be selected for the sample can be specified (e.g. Random sample) ○ When describing the behavior of a particular population ○ A representative sample - we can draw accurate, unbiased estimates of the characteristics of a larger population ○ It has to have: External validity - the sample has to reflect the population Internal validity - when we assign certain treatments to certain groups (random assignment), we have to be sure the results in our data are actually thanks to the different treatements and not differences between the people in those groups ○ Sampling error - it always differs from the actual characteristics of the general population Using the error of estimation to determine how the data obtained is expected to deviate from the population as a whole Affected by the sample size, population size and variance of the data ○ Simple random sample - done by sampling frame (list of population from which the sample will be taken) Then use tables of random numbers to select participants ○ Systematic sampling - taking a specific number of individuals for the sample (e.g. Every eight person) -> after selecting an individual, several others do not have the same chance of being picked ○ Stratified random sampling Dividing population into two or more subgroups or stratum (subset of population sharing a specific characteristic - e.g. Age, race,…) and then random sampling from these stratas (Sometimes) Proportionate sampling method - cases are sampled from each stratum in proportion to their prevalence in the population ○ Cluster sampling - we cannot always obtain a list of everyone Dividing into clusters (usually by geographical factors or particular situations, not by a certain characteristic like age, gender,…) Usually done in a multistage manner - larger clusters divided into smaller clusters ○ The nonresponse problem - failure to obtain responses from selected individuals Researchers increase the amount of people in the sample, contact the non responsive people multiple times Offer incentives as a form of payment Designing simple studies, that are fast, offering more languages Tell people in advance, rather than random contact them  Do the respondents and nonrespondents differ in a systematic way? - yes/no ○ Mis-generalization When researchers draws conclusions about different populations than from which the sample was taken Nonprobability samples - testing how variables relate to one another, things like error of estimation do not apply ○ Convenience sampling Testing people that are already available (people in a clinic, store, on the street) ○ Quota sampling Certain kinds of participants are obtained in particular proportions (aka 20 men and 20 women rather than 40 random people) ○ Purposive sampling Using past research findings to decide which participants to unclose + using researcher's own judgement Choosing the respondents who are typical to the population they want to study E.g. Past on previous elections - determine areas of voters and predict the upcome of the next election But not reliable (we are using the researchers judgement) How many participants? ○ Error of estimation - we have a certain number we want to achieve -> calculate how many people we should need ○ Using economic samples - reasonably accurate estimate of population at a reasonable effort and cost ○ Power = the ability of research design to detect any effects of the variables being studied in the existing data (higher powers detects more effects) Increases with the sample size, some strong effects can be seen even with a smaller sample Two main types of statistics that make use of data collected: ○ Descriptive statistics Exloratory data analysis (EDA) - throughoutly examining data in detail ○ Inferential statistics - using data from samples and applying it to the whole population A measure reffering to an entire population = parameter (when calculated from a sample of data = statistics) Parameters are the real entities of interest Descriptive research - describing characteristocs/behavior of a given population in a systematic and accurate fashion ○ Not designed to test hypotheses, just used to provide information ○ A good description is accurate, concise and comprehensible ○ Types: Survey research Respondents provide information about themselves Cross-sectional survey design - a single group of respondents is surveyed Can provide info about how certain groups differ Successive independent sample survey design - two or more sample groups answer the same questions at different times of their life (e.g. How many people go to church asked in 1950s and 2000s) But we always have to make sure the samples are comparable Longitudinal/panel survey design - a single group is questioned mote than once But problems in not getting all the participants from before (some died, moved, dropped out) Demographic Research Describing and understanding patterns of basic life events and experiences (marriage, divorce, employment,..) Used to rofecast changes in society that might require government attention, understanding why people have the amount of kids they do,… Epidemiological Research Occurrence of disease and death in different groups Psychologists - study the behavior of people with an illness -> able to target protentional people at risk + study the prevalence (proportion of population that has the disease at a particular point in time) and incidence (rate at which new cases occur over a specified period) ○ Always needs to be presented and described Data should be accurate, concise and understandable But the most accurate description of a set of data is raw data (which can be overwhelming) and more concise versions might distort the results Researchers chose selectively which data is gonna be presented to show the result in the clearest way possible Can be done in a numerical or graphical way Frequency distributions  = table that summarizes raw data by showing the numbers of scores that fall in certain categories Aka organizing data in a logical order Abosolute frequencies (f) Number of respondents with a given score Hard to interpret and compare Cumulative absolute frequency (only for ordinal, interval and ration methods of measurements) - all observations of previous categories are added to the current category Simple frequency distribution - indicates the number of participants who obtained each score, arranged from lowest to highest and then the frequency of each score is shown - not the best for ratio and and interval m.m. (i have a lot of dataú -> Grouped frequency distributions - frequency of subsets (class intervals) of scores Intervals are mutually exclusive, capture all possible responses and are of the same size (e.g. People aged from 1-5, 6-10, etc.) Sometimes include relative frequency (P, P=f/N) as well (the proportion of the total number of scores that falls in each class interval, all relative frequencies add up to 1) Cumulative relative frequencies F (only for ordinal, interval and ration methods of measurements) - all proportions pf previous categories are added to the proportion of the current category True limits of an interval (called the real lower limit and the real upper limit) - never whole numbers They should always include the smallest/highest value that would be clased as falling into the interval E.g. Instead of the interval being just 35 - 39, it includes all values between 34.5 and 39.5) Frequency histograms and polygons Sometimes it's just easier to show it on a graph  Histogram - variable on the x-axis is on an interval/ratio scale (bars on the graph can touch each other), vizualisation of a group frequency - suitable for all scales Bar graph - variable on the x-axis is on a nominal/ordinal scale (bars are separated to not indicate that the variable is continuous) Equal differences in scale values to not reflect equal differences in characteristic being measured) Bar chart/pie chart - nominal and ordinal scales Polygon - lines connect the frequencies in class intervals Central tendency Summarizing the data of the entire group of participants Provide info about the most typical score Done by measures of central tendency (or measures of location) The mean - the average Influenced by extremes Can be manipulated algebraically May be an indication of a more stable central tendency (if we calculate more means from many samples) Trimmed means - calculated on data for which we discarded a certain percentage at each end of the distribution Works to trim the extremes and avoid skewness The median (Mdn)- less affected by extremes (outliers) Median location - to see which number is the "line" is the median The mode (Mo) - score obtained from the largest number of subjects The only score that always occurs ○ Presenting means in tables and graphs  Error bars - provide information about the researchers confidence of each mean (since the mean calculated is always only an assumption) How correctly have we calculated our mean? -> confidence interval (CI) - when subtracted or added we get the range of scores through which the mean can generally pass ○ Measures of variability/dispersion If small - the mean can correctly assume the typical participant's score Range + variance + standard deviation The interquartile range (again, to get rid of extreme scores) - disregarding the upper (point of cut off - third quartile Q3) and lower (point of cut off - first quartile Q1) 25% Q3-Q1= interquartile range When using a trimmed sample - using a Winsorized sample Dropping a certain percentage of the lowest scores and replacing them with the copies of the smallest score that remains and the same thing for the highest scores Mean absolute deviation - averaging all the deviations using their absolute value, summing them and dividing by N Most variables fall into the normal distribution, but we can also have positively skewed distribution (less high scores) and negatively skewed distribution (more high scores) Cca 68% of the cores will fall in the range defined by +-1 standard deviation Sample variance Very sensitive to extreme scores Comparing standard deviations on measures with different means - we have to scale the s.d. by the magnitude of the mean -> COEFFICIENT OF VARIATION ○ ○ But keep in mind that if the scale is arbitrary - variance and coefficient might not be reliable enough  Distinction between biased and unbiased estimators ○ Expected value - the long range average of many samples We want it to be equal to the population mean (or any parameter) that it is estimating -> unbiased estimator (both the sample mean and sample variance are unbiased estimators) BOXPLOTS ○ For interval, ratio and ordinal ○ Using the quartile location (aka like the median location for median but for quartiles) ○ ○ Inner fences - point that falls 1.5 times the interquartile range below or above the appropriate quartile ○ Adjacent values - actual values in the data that are no more extreme (farther from the median) than the inner fences ○ Used when we want to compare several groups ○ E.g. Data and calculations + boxplot How to describe a distribution? ○ Overall pattern Number of peaks? Symmetrical/skewed? Tails think/thin? - aka how far to the sides does it go Central tendency - midpoint Spread - a little of a lot ○ Deviation from the pattern Outliers ○ Cavecats (aka misleading numbers and graphs) Omitting tje baseline Manipulating axes Cherry picking data Using the wrong graph Going against conventions  Histograms sometimes fail as a clear description of data (especially smaller sample sizes) -> we use fitting curves (interval or numerical) = ideal approximation of emipirical distribution ○ Fitting a normal curve - we often assume that our data are normally distributed We fit a normal distribution on our histogram and see how good our assumption is (aka is it really highest in the middle or does it slightly lean towards one side) Total area under curve = 1 ○ The kernel density plot They try to fit a smooth curve to the data while at the same time taking account of the fact that there is a lot of random noise in he observation that should not be allowed to disturbed the curve (pay no attention to the mean and standard deviation) We have to realize that each observation might have been a bit different than recorded We use the recorded data and put it on the x-axis as a distribution (for example a normal one) We do this for all data and then notice where they overlap -> vertically add the overlapping values to create the curve Describing distributions ○ Symmetric - those that peak generally around the center ○ Bimodal - has two peaks (any distribution that has two predominant peaks) ○ Unimodal - only one peak ○ Negatively/positively skewed ○ Kurtosis - refers to the relative concentration of the scores in the center, upper and lower tails and the shoulders of a distribution Mesokurtic - tails are not too thin or too thick Platykurtic - distribution becomes flatter  Leptokurtic - more peaked Normal distribution ○ We frequently assume normal distribution in the population ○ Assuming a variable is normally distributed -> allows us to make a number of inferences ○ Abscissa - horizontal axis (different possible values of Xi), ordinate - vertical axis (the density. Related to the frequency/probability of occurrence of X) ○ ○ Trying to table normal distribution - problem (we would need to make a different table for every possible combination of values for the mean and std.d. when we want to compare different scales, scores are hard ti interoret without context) -> standard normal distribution Mean = 0, std.d. = 1 (they become parameters) We are able to move the score along the X-axis Standard deviation become 1/sigma times larger (if its bigger than one -> we have to make the distribution narrower and vise versa) When we want to figure out how the individual will get a score above some particular value -> transform it to a score of z (that comes from the standard normal distribution) z=pivotal quantity, its distribution does not depend on the values of the mean and std.d. represents the number of standard deviations that Xi is above or below the mean This transformation does not affect the shape of the distribution (if it wasn't normal before, it won't be now) Determining probabilities (proportions) based on scores ○ Formulate question and draw a picture ○ Standartize (z scores)  ○ Estimate the proportion F with the 68.85-95-99.7% rule (to really get a sense of the distribution) 68% of distribution lies no more than one std.d. from the mean, etc. Formulate a conclusion Types of probabilities (using the cumulative relative frequencies) ○ Less than: P (X < x) ○ Greater than: P (X >= x) - cumulative relative frequency substracted from one ○ Between: P (x1 then we compare both quantiles Put them on a graph and if they form a straight 45 degree line -> the distribution is normal  ○ Correlational research ○ Correlation coefficient Indicates the degree to which two variables are related to one another in linear fasion Mostly used - Pearson correlation coefficient - r Ranges between -1 and +1 Standartized, does not get influences by unit of measurement Variables might be positively or negatively correlated Positive - direct, scores of one variable increase the scoes of the other variable Negative - inverse, if score of one increases - the other decreases The higher the number, the stronger the correlation (without the + or -) ○ Graphic representation In scatter plots Positive cor. - upwards slope to the right Negative cor. - downwards slope to the right The stronger the correlation - the more clustered he data is (perfect one would make a straight line) Predictor variable - x axis, criterion - y axis Regression lines - given any specific value of X, the corresponding height of the regression line represents our best possible prediction of Y - how close points are to the line indicates the correlation Sometimes variables are not linearly related but curvilinearly related Can be also homogenous - dots in a single cloud, or heterogenous - dots form multiple clusters  ○ Coefficient of determination Helps us understand what the number of r itself actually tells us Squared r - tells us the proportion of variance in one of our variables that is accounted for by the other variable (indicates the proportion of the total variance in one variable that is systematic variance shared with the other variable) Basically how to know what is actually correlation and what is just by chance Covariance - the degree to which two variables vary together, not standartized - gets influenced by the unit of measurement ○ But it is not an unbiased estimate of the correlation coefficient in the population (rho) -> using the adjusted correlation coefficient ○ Statistical significance of r Exists when a correlation coeffcient calculated on a sample has a very low probability of being zero in the population Affected by three factors Sample size bigger the better Magnitude of correlation The further away it is from 0 the better How careful we want to be not to draw an incorrect conclusion about whether the correlation we obtain could be zero Using tables to determine Have to know sample size Absolute value of r Whether we have made a directional or nondirectional hypothesis Directional - predicts the direction of the correlation (whether it is positive or negative) Nondirectional - only predicts that two are connected but not how ○ Factors distorting correlation coefficients Restricted range  When i only have scores from a very homogenous group -> i can be mislead into believing there is weak correlation Also affects it when we have curvilinear correlation - i only see a part of the curve Usually reduces correlation - we have to make sure that the coefficient is not innapropriate for the question at hand Outliers On-line outliers - they still look like a part of the pattern but they are extreme -> make correlation bigger than it is Off-line outliers - artificially deflate the value of r Reliability of measures If we have unreliable ones -> affect the magnitude of correlation coefficients Heteregenous subsamples - when using data from different sources (it might be affected by an unaccounted for third variable) ○ Correlation and causality Relationship between two variables is either: Association (interdependence) Both variables are related No distinction between independent/dependent variable Dependence Prediction (regression) or causality Independent variably X predicts dependent variable Y Correlation does not imply causality To say one variable causes another one: Covariation Changes in one variable should change the values in the other variable (aka correlation) Directionality Must show that the presumed cause precedes the presumed effect in time (correlation studies variables at the same time) Elimination All extraneous factors that might influence the relationship must be eliminated (cor. Research can never fulfill this fully, what if there is another third variable responsible for both of these two variables?) Allows us to draw conclusions The third variable problem ○ Confounding variable = a variable that influences the interpretation of the relationship between other variables Called Z Consequences: Stronger/weaker than in reality In the opposite direction Based on spurious relationship -> a true relationship becomes invisible We need to keep them under control - thanks to experimentation A well designed experiment has three properties ○ One independent variable is varied to assess its effects on the participants ○ Participants are divided into groups in a way that ensures their initial equivalence ○ All extraneous variables that might influence the responses are controlled Manipulating an independent variable ○ An independennt variable has two or more leves (aka values) One group gets two candies, the other four, the third six -> three levels Also refferred to as conditions Sometimes involve quantitative differences (aka the levels differ by quantity), other time qualitative differences (different instructions) In control conditions - offers a baseline, placebos ○ Types: Environmental manipulations Modifications of aspects of the research setting Can be used in social situations as well - using confederates (accoplices of the researcher) to alter the conditions Instructional manipulations Modification through information that participants receive Invasive manipulations Creating physical changes in the participant's body through physical stimulations, surgery, drugs Experimental and control groups ○ One level of the variable might involve the abcense of the variable of interest ○ Experimental groups - have some level of the independent variable, they are qualitative or quantitiative, control groups have zero  ○ Control groups are used when the researcher wants to know the baseline level of a behavior in the absence of the ind.v. Assessing the impact of ind.v. ○ Pilot testing the levels we plan to use - to see if the levels are different enough to be detected by participants ○ Manipulation checks - questions that is defined to determine wheteher the ind.v. was manipulated successfully (usually after) Subject variables ○ Differences in participants that are not experimentally manipulated by the researcher ○ They reflect the existing characteristic of the participants ○ No manipulation possible but they still have an influence ○ -> quasi-experimental conditions Dependent variables ○ The response being measured in the study Assigning participants ○ We want to make sure participants in our groups did not differ before the experiment began ○ Simple random assignment - every participant has an equal probability of being placed in any experimental condition ○ Blocked random assignment - two groups, assign half on one group to one condition and half to other condition, then the same with the other group ○ Matched random assignment - obtaining participants' scores on a measure known to be relevant to the outcome of the experiment -> ranking them based on it -> putting participants in clusters of size n, (n=number of conditions in an experiment) -> random assignment of n participants in each cluster to each of the conditions ○ Repeated measures assignment Normally we use the randomized groups design (aka between-subjects design - we are interested in differences in behavior between the groups, for those we use the previusly mentioned methods) Within-subject designs - one group is tested for all the conditions -> repeated measures design No need for random assingnment More powerful - participants are identical in every way, differences are just thanks to changes in the independent variable Require fewer participants Order effects - when a response is affected by the order in which they participate in the conditions Practice  Performance improves because they complete the dependent variable several times Fatigue Sensitization Participants realize what the hypothesis is and start responding differently Using counterbalancing - using different orders for different participants Latin square design - each condition appears once at each ordinal position and each condition precedes and follows every other condition once Carryover effects - effects of a condition persist even after the condition ends The new behavior us due to the lingering level of the independent variable presented earlier Experimental control ○ Systematic variance revisited is any of the total variability we observe in the score systematic variance due to the independent variable? We should observe systematic differences between the scores in the various experimental conditions Can come from two sources: Treatment variance - portion of he variance in participants' scores that is due to the independent variable Cofound variance - variable other than the independent variable has an effect Portion of the variance that is due to extraneous variables that differ systematically Must be eliminated Internal validity - the degree to which a researcher draws accurate conclusions about the effects of the independent variable Biased assignment of participants to conditions Effects are due to nonequivalent groups Attrition Loss of participants Different attrition (loss of participants during the study) When the level of attrition differs across the experimental  conditions (how do we know then that differences are due to dependable variable or a characteristic of the participants?) Pretest sensitization Subject reacting differently because they have already underwent a pretest History Extraneous events that occurd outside of the research setting - reaction between the independent variable and hisory effects Something from the past can be reminded leading to a change in attitude Miscellaneous design confounds When some participants are treated differently Maturation of participants over time Issues with instrumentation - aka the measuerement instrument Experimenter expectancies, demand characteristics and placebo effects The validity of researchers' interpretations of the result of a study are affected by beliefs as well Experimenter expectancy effects Researchers have an idea on how participants will respond - risk of seeing what we want to see Demand characteristics Participants' assumption about the nature of a study Aspects of a study that indicate to the participants how they should behave Using double-blind procedures to help - neither the participant nor the researcher interacting  knows the true nature of an experiment Placebo effects Physiological or psychological change that occurs as a result of the mere suggestion that the change will occur Error variance ○ Unsystematic differences between people ○ Does not invalidate experiments - we have statistical ways to distinguish between treatment variance and error variance ○ Sources Individual differences between participants We are all different and might respond differently to certain situations Using homogenous sample of participants - less error variance is produces by their differences Transient states The active state of oneself during the study (mood, illness,…) Not changing a researchers attitude, remaining the same towards everyone Environmental factors External noise, weather, different times of the day \ Trying to hold a constant environment for all participants Differential treatment Moods and health of researchers can also affect the way they act towards participants Automatization of the experiment as much as possible - using computers, programmed equipment Measurement error Tight experimental control leads to an often artificial situations -> harder to generalize the findings ○ External validity - the degree to which the results obtained in one study can be replicated or generalized to other samples, research findings, etc.  Experimenter's dilemma - high internal validity leads to low external validity The purpose of most experiments is not to discover what people in real-life settings do (finding of any single experiment should never be generalized) The purpose of experimentation is to test general propositions Web-based experimental research ○ Obtaining much larger samples with lower expenditure of time and money ○ Samples ae more likely to be diverse ○ It is easy to obtain samples with specific characteristics ○ No researcher present - data obtained may be less affected by biases ○ Difficulty identifying and controlling the nature of a sample ○ Cannot control the setting to minimize error variance ○ Frequent failure of participants to fully complete a study ○ Limited in the research paradigms that may be used Experimental design ○ One way design Only one independent variable is manipulated Two group experimental design Minimum of Two conditions to compare participants responses Three basic varieties: Randomized groups design Participants are randomely assigned to one of two or more conditions Matched subjects design To increase similarity Participants matched into blocks on the basis of a variable the researcher believes relevant to the experiment Repeated measures/within-subject design Each participants serves in all experimental conditions Posttest-only design - the dependent variable is measure after the experimental manipulation has occurred Pretest-posttest design - dependant variable measured twice (before and after manipulation) Helps to verify that participants in the varioys experimental conditions did not differ in the beginning of the experiment See how mucg the independent variable changes the behavior  More powerful - more likely to detect the effects of the ind.v. on the dependent variable Error variance due to preexisting diffeences is eliminated Can lead to pretest sensitization or providing info about the nature of the experiment ○ Factorial design Two or more ind.v. are manipulated (reffered to as factors) To study the individual and combined effects of two or more independent variables within a single experiment How do we describe the size and structure of such designs? Described in a way tat immediatelu indicates to a reader how many independent variables were manipulated and how many levels were there for each of them A 2 x 2 experiment (two independent variables of two levels), a 3 x 3 (two ind.v. of three levels), a 2 x 2 x 4 (three ind.v., two of them have two levels and one has four), etc. -> we figure out the number of possible experimental conditions by mulitplying all numbers Assigning participants Randomized groups factorial design Random assignment to one of the possible combinations of the independent variables Matched factorial design Matching participants into block on the basis of some variable that correlates with the dependent variable - as many participants in each block as there is experimental conditions Then participants in each block randomely assigned to one of the conditions Repeated measures factorial design All participants in all conditions When a lot of conditions - kinda an issue Mixed factorial design Also called the between-within design or split-plot design Basically using different methods for different variables Provide info not only about the separate effects but also of the combination of the ind.v. Examine whether the variability in scores was due: Individual effects of each ind.v. To the combined or interacrive effects of the ind.v.  Error variance Main effects = The effect of a single independent variable Ignores effect of the other ind.v. As many effects as there are ind.v. (e.g. In a 2 x 2 x 3 design, there will be three) Interactions Present when the effect of one ind.v. differs across the levels of other ind.v. We say variavles interacted when the outcome of a behavior is different when both are present ○ Combining independent and participand variales Subject/participant variables = sex, age, ability, attitudes,… Designs, where the indepdendent variables rae manipulated and the participants variables are measured -> expericorr factorial designs Investigating the generality of an ind.v. effect Do the effects of an ind.v. affect everyone or just participants with certain attributes? How do certain characteristics relate to behavior under varying conditions? Classifying participants into groups Discrete participant variables (gender, political affiliation, race) - using groups based on these variables and then random asg. Continuous variables (self-esteem, level of depression,…) - median-split procedures A median is identified in the scores Classify subjects that are below and over the median Or using extreme groups procedure Oretesting a large number of pa=otentional participants, selecting participants with unusually low or high scores But generally should not be used - we are only seperating subjects into two groups - we lose information about the variability in scores Leads to bias and missing effects that are actually present We should rather use multiple regression procedures that allow us to analyze data Interpreting redults  We can draw causal inferences only about the ind.v. that were actually manipulated We can only say the subject variable moderates the reactions to the ind.v. Does not involve random assignment of participants to conditions In many instances, the researcher doesn't manipulate the independent variable at all ○ Quasi-independent variable - an event that participants experience for some reason Do not posses high internal validity, but still tries to eliminate many of the threats to internal validity ○ Pretest-posttest designs Never use just one group (aka using O1 X O2 design - observation number one X observation number two) Many possible factors affecting the outcome Regression of the mean - the tendency of extreme scores in a set of data to move or regress toward the mean of the distribution with the repeated testing In the pretest, there are many factors that can contribute to the extremness of scores that are not present in the post test Called the preexperimental deisgn - lacks control, no internal validity Having more control groups A true control group is not possible, but we can have nonequivalent control group designs Having one or more groups of participants that appear to be reasonably similar to the grop that received the quasi-independent variable Two varieties Posttest- only ○ Measuring both groups after one of them has received the quasi-experimental treatment ○ But we don't know if the differences are not thanks to preexisting ones Pretest-posttests design ○ Have the groups scored similarly on the dependent variable before the introduction of the treatment? ○ But still some setbacks - like the local history effect (one  event occuring that did not occur to the other group) - the selection by history interaction How do we make sure our groups are as similar as possible? Collecting dara and information to explore the possible differences ○ Time series designs Measure the dependent variable on several occasions before and on several occasions after the quasi-independent variable occurs Simple interrupted time series design Several pretest measures are taken before introducing the variable and then taking several posttest measures afterwards Repeated measures of the dependent variable have been interrupted by the occirrence of the quasi-independent variable We should be able to distinguish changes due to aging or maturatuon Problems with contemporary history - the observed effects being due to another event that occures at the same time as the quasi-independent variable Interrupted time series with a reversal What happens to the behavior if the quasi-independent variable is introduced and then removed We are interested if removing the variable will lead to the behavior returning to the pre-variable level Problems Researchers often do not have the power to remove the variable The effects of some variables remain even after the variable itself is removed The removal may produce canges that are not due to the effects of the variable Control group interrupted time series design Helps to rule out some history effects ○ Comparative time series design Examines two or more variables over time in order to understand how changes in one variable are related to changes in another variable Provides indirect evidence that one variable might be causing the other one ○ Longitudinal designs  The quasi-independent variable is time itself Usually to study age related chnages Goal is to uncover developmental changes that occur as a function of age But something other than age related development might produce the change Problems Reserachers find it difficult to obtain samples of participants who agree to be studied agaian and again Trouble keeping track of the participants It is time consuming and expensive Why use it instead of cross-sectional designs? They cannot dustinguish are related changes from generational effects They allow researcher to examine how individual participants change with age ○ Cross-sequential cohort designs Allows to rease apart age and cohort effects Two or more age cohorts are measured at two or more times ○ Program evaluation Behavioral research methods used to assess the effects of interventions or programs designed to influence behavior Their goal is to provide information to those who must make decisions about the target programs ○ Evaluating Quasi-experimental designs Meets two out of the three criteria for infering causality - the presumed causal variable preceded the effect in time + the cause and effect covary But - cannot fully eliminate the effects of extraneous variables We want to combat as may threats to internal validty as possible Meassure not onl the effects of the quasi-independent variable but also the processes that are assumed to mediate their relationship - provides more conficence Using critical multiplism - critically considering many ways of obtaining evidence relevant to a particular hypothesis and then employing seveal different approaches Some threats arise when examine a single group of participants, some appear when we compare groups  Single-case research ○ Debating which approach is better in behhavioral science: Nomothetic approach - seeking to establish general principles and broad generalization that apply across individuals But problems with exceptions - it does not apply to everyone Idiographic approach - seeks to describe, analyze and compare the behavior of individual participants We should focus not on the general trends but also on the unique behaviors of specific individuals -> using single-case experimental designs Researchers manipulate independetn variables and exercise strong experimentla control over extraneous variables + analyzing the behavior of individual participants rather that the grouped data -> using case studies - behavior of an individual is described in detail Involve uncontrolled impressuonistic descriptions ratger than controlled experimentation ○ Single case experimental designs Group designs refelct the most common approach Criticism: fail to handle: error variance - averaging the responses of several participants shouls provide a more accurate estimate of the typical effect on the independent variable We can also estimate the amount of error variance in our data BUT - much of error variance in data does not reflect variability in behavior  but rather is created by the group design itself And researchers accept the presence too blithely - much of error variance is due to individual differences among participants (interparticipant variance) which they are not concerned with In single-case studies, we are interested in the intraparticipants variance Generalizability - adding all the scores together reduces the impact of the idiosyncratic responses of any particular participant (they can identify the overall effect of the independent variable) But the average might not accurately represent the response of any particular participants Generalizations might be misleading reliability - they demonstrate the effect of the ind.v. a single time (if they do a replication, they do it in later studies) Single case - using intraparticipant replication (introducing an ind.v., removing it and then again add it) + interparticipant replication - studying the effects of the ind.v. on more than one participants Single case experimental design - more participants, but their responses are analyzed seperately Data is not analyzed using inferential statistics Usually presenting data with graphs that show results individually for each participant - using graphic analysis We can visually see if the variable had any effect Only the strongest effects are accepted as real ABA designs Attempt to demonstrate that an ind.v. affects behavior thanks to showing it affecst a behavior and taking it away to show it causes the behavior to cease Includes a baseline period (A), followed by introduction of the ind.v. (B), followed by reversal period when the ind.v. is removed (A)  Sometimes the ind.v. produces permanent changes - we cannot see if the initial change was due to the variable or not But if it happens with multiple participants - we can be pretty confident Multiple-I designs Testing differences among levels of ind.v. Obtaining baseline, introducing one level of the variable, then removing it and trying another level, etc. Sometimes adding another baseline period - usually to assess the effects of drugs Multiple baseline designs Two or more behaviors are studied simultaneously An ind.v. is introduces that is hypothesiyed to affect only one of the behaviors When are they useful? Usually to study operant conditioning + for studying psychophysiological processes + sensation and perception Effects of behavior modification - changing problem behaviors Critique Do not really possess good external validity Not suited for studying interactions among variables Ethical issues with using ABA - is it chill to withdrawl something that could be potentionally helping just to see if it actually IS helping?? ○ Case study research Detailed stidy of a single individual, group, or event Adding data together into a narrative description of the particular person, event,.. Sometimes uses subjective impressions of the researcher - using available infromation to interpret and explain how and why the individual behaved as they did Source of insights and ideas when in early stages of investigating a topic Describing rare phenomena Psychobiography - applying concepts and theories from pscyhology in an effort to underdtand the lives of famous people Illustrative anecdotes - using case studies to show an example rather than using an abstract one Limitations and critiques Failure to control extraneous variables  Useless to provide evidence to test behavioral theories or psychological treatments Unable to assess the viability of alternative expanations of their observations Observer biases Rely on the observations of a single researcher Risk of self fulfillng prophecies and demand characteristics