Modules (1-3) Flashcards

Question

what is regression to the mean?

Answer 1

- Any group at the extremes of a distribution will shift to mean over time. - To improve players batting averages he offers those with the lowest average to donate money to charity on every run they score. Their batting average increases with the incentive. - He tried the same thing with his best players an found that their averages fell. He concludes that they do not like charity. Why is this faulty logic? Regression to the mean, statistical fact, if we select people at the extreme of the distribution and look at them over time they will regress to the mean (worst get better and best get worse). People fluctuate over time on traits but will always come back to the mean/average naturally (compared to self).

Answer 2

- Is unlikely to equate groups on all extraneous variables - Two participants (100-120) - Ten participants (100, 120…) - The mean will be much higher differences between conditions with smaller sample size relative to larger sample sizes. The more people there are in the group the more likely it will fall into a normal distribution and that we are sampling close to the mean. Smaller sample sizes make us more likely to be distributing from extreme ends of the bell curve and skew the data. - Random assignment is designed to create equal groups on extraneous variables. Small samples are not likely to be similar to each other even with random assignment! - How large is large? It depends on how variable about the thing you are measure/control for. A good rule of thumb is 30 people in each group (central limit theorem-normal distribution-sampling from population to get close to the real mean) - One trick is to sample from people who are already similar with one another. - How many people does it take to equate groups on anxiety? • If you selected from the NZ population? o Huge variability. • If you selected from first year PSYC? o Now I have people with similar interests, education level, stress etc. may not need as many people in each group because there is less variability. • If you selected from a group who are prone to anxiety (e.g., family members of people with an anxiety disorder)? o Equated genetic risk factor. • If you selected from people who have clinical anxiety? o Very small level of variability within participants and will need an even smaller sample size. - How small to go when sampling is a question of having Homogeneous vs heterogeneous samples? - Do you want a sample very similar that’s homogeneous with a smaller sample size which can make it les representative of the wider population. - Do you want a variable sample with a larger sample size that is representative of the wider population. - = external validity.

Answer 3

``` - Treatment response in patients given an “inert” substance or treatment. - Psychological improvement - Physiological improvement - Influenced by context ```

Answer 4

``` - Side effects in participants given an “inert” substance or treatment. - Common symptoms - Warnings - Ethical? ```

Answer 5

- A series of productivity studies at Hawthorne Electrical works - Changed lighting, equipment, work schedules, teams etc. - Most changes led to increased productivity as long as they were noticeable ii. Workers building lightbulbs in a factory to test whether organizational changes would increase productivity. They found no matter what they did it always increased work productivity. They were smart and noticed these changes and guessed they were installed to increase productivity, so they changed their behaviour to meet researchers expectations. The change in DV is not due to the IV but researcher expectations - The Good Subject (Orne, 1962) - Most research participants aim to please - Obedient to a fault - E.g., a hypnosis study, that people weren't hypnotized but just following instructions. He did a study without hypnosis and gave them an insane task to do, mind numbingly boring and tedious task, they all did as they were told. They respond to the perceived authority of the experimenter. Social desirability effect they want to be viewed in a positive light by the researcher, so they are not as truthful or change their behaviour. - Social Desirability - Most research participants want to be seen in a positive light - May not be truthful or accurate in admitting negative thoughts/behaviours Solutions: 1. Well designed control condition[s] 2. Participants are blind to condition/treatment 3. Participants are blind to hypothesis 4. Obscuring purpose of methods/scales/measures (deception; filler tasks; confidential/anonymous) 5. Manipulation checks (test after study if they knew the hypothesis) 6. Debriefing (what do you think the study was about? What were your expectations)

Answer 6

- Sometimes experimenters treat participants differently depending on condition. - During data collection - During data coding - Rosenthal & Fode (1963) - Student experimenters. - Tested Maze-Bright and Maze-Dull rats. - Rats were randomly assigned to be bright or dull. - Genetic breeding study on rats, create rats that were smart enough to go through maze (bright or dull rats). It was a deception, all were the same and counterbalanced, but student researchers were told some were smart and others were dull. This influenced how the experimenter interacted with the rats. - Results: - Still “bright” do better than “dull” - Gets taller overtime to show they’re learning/getting better with practice. - Rats are treated differently, handling, encouragement, setting up maze, where they start (subtle unsconious bias in behaviour in favour of bright rats from experimenter). - Rosenthal & Jacobson (1968) - Elementary students given a “new” IQ test at beginning of school year. - 20% randomly identified as “late bloomers”. - Given standard IQ test at year end. - Same study on humans: - The self-fulfilling prophecy effect. - They randomly labeled some kids as late bloomers, poor now but will boost soon in IQ, others are controls. The late bloomers did better because they had better interactions, support, encouragement from teachers relative to controls (expectations of students impacts on their performance; +/- effects on students, unconsciously our expectations come though in our behaviour). - Or generally treat people to make them more likely to give the desired outcome we want/expect. Solutions: • Experimenters blind to condition • Double blind – experimenters and participants are both blind • Experimenters blind to hypotheses (i.e., undergraduate research assistants who do not know hypothesis till the end of the study). • Automated data collection (computer doesn’t have expectations! Reduce experimenter interaction and bias).

Answer 7

- After data is collected the experimenter can still bias the results. - Subjective Coding can occur when the DV requires human judgment (room for error, bias or expectations) - In observational studies or developmental research. - Solution: • Automate scoring (time, speech sample and analysis etc.) • scoring rubric (set rubric each scorer uses, standardise it) • Inter-rater reliability (multiple independent raters with 80% at least agreement) • Coders blind to the condition/hypothesis • Different data collectors to coders!

Answer 8

``` Type I error = α (Greek letter alpha). • Rejecting the null when it is actually true. • False positive • Innocent man is imprisioned. ``` ``` Type II error =β (Greek letter beta). • Failure to reject the null when it is actually false. • False negative • A guilty man is set free. ``` There is always a chance for error! Therefore, they set predetermined criteria for what constitutes as sufficient evidence to reject the null hypothesis. The alpha (p-value) are the amount of evidence required to reject the null hypothesis

Answer 9

Whether I can conclude that there is sufficient evidence to attribute the found difference to the manipulation of the IV?

Answer 10

All test statistics are a measure of the variability that is due to your effect, compared to the variability that occurs in your population generally.

Answer 11

SEM is used to estimate how close the sample mean is to the true population mean.  The larger the sample we have the more confident we are that it will be closer to the true mean. There will be variability in the mean obtained with every sample, how spread are these means?  SEM = S / square root of the sample size to scale it to fit the population  The smaller the SEM the more confident we are closer to the true mean.  As we increase the N, the SD becomes more accurate (closer to the true mean)  As we increase the N, SEM becomes smaller.  The mean we get in sample will always vary when we sample, we want it as close to the true mean as possible (need 30 per condition or smaller if it’s a within subjects design).

Answer 12

o The likelihood (or probability) of obtaining the observed result (or a result more extreme) if the null hypothesis is true. (i.e., less than 5% chance that the results are due to chance or a false positive error; if the p-value we obtain is below 0.05 then we reject the null hypothesis because it is very unlikely that we are making a type 1 error). o The likelihood that two samples are drawn from the same population. o By convention in psychology, p < .05 is significant .05 is our alpha

Answer 13

 In the population, people’s RTs should be normally distributed (some people will be very fast, some people will be very slow, most people will be in the middle).  If I randomly select 30 of these people and put them in the angry group, and randomly select 30 and put them in the happy group, there is a probability that the two groups won’t have EXACTLY the same mean RT.  So, there is a difference between my groups before I’ve even done anything to them. EVEN WITH RANDOM ASSIGNMENT. This variability of the observed difference compared to the true difference is called SAMPLING ERROR (i.e., natural variation of the mean which occurs when sampling from the population). Sampling Error & the T Test Statistic:  Sampling error impacts the t-test score significantly! If the null hypothesis is true and there is no difference between the groups mean.  The T-test score will never be 0 because of sampling error (variability between the sample mean and true mean with each sample).  t = 0 - No difference between groups (they are drawn from the same population) - X1 – X2 = 0  Sampling error is chance; will sometimes be bigger or smaller than but never 0 (even if there is no difference between two group means). *Main Q is weighing up the variance explained by the IV and noise (random error)

Answer 14

 Sampling distribution shows me how often I will get different values of a test statistic (e.g., t) by randomly sampling from a single population if the null hypothesis is true.  Shape of the sampling distribution depends on how big my sample is (degrees of freedom = N – 2; 1 for each group)  Shape of the t distribution depends on the number of people in your sample. The more people, the more likely t will be close to 0.  Red = small degrees of freedom of 3 (shallow with more spread; variety of t’s in small sample)  The red ends of the tail is the tiny part of the distribution where in this area (less than 0.05%) will these findings be due to chance (extreme values +/- that can only occur 5% of the time; 2.5% either side).  The larger the sample size and degrees of freedom the steeper it gets, and the less likely extreme values are = closer to true mean! Random differences between groups not due to IV is less likely.

Answer 15

 If my t is very big, then it is very unlikely my two means come from the same population. So I will assume that I have caused my two groups to have different response times through my manipulation (my manipulation is guilty).  Critical value of t - Value at which there is less than 5% probability that the t arose through sampling error. - We therefore reject the null hypothesis. - We conclude that there is a significant difference between groups. *if it meets the critical threshold of t (at p-value of 0.05; sampling error) we can reject the null hypothesis because the findings are not likely to be due to random chance and is attributable to the manipulation of the IV to the difference found.

Answer 16

 H0 = no difference between RT to angry and happy faces  H1 = RT to angry faces are different than RT to happy faces - Bidirectional hypothesis (difference in either direction, test both tails, 2.5% of tail at both ends) - Two-tailed  H1 = RT to angry faces faster than RT to happy faces - Directional hypothesis (one group better than the other, test one tail; 5% at one end) - One-tailed