Ethics of Research - Review and Application + Some “Catch-up” Items Lawrence R. Gordon Psychology Research Methods I.

Ethics of Research - Review and Application

+ Some “Catch-up” Items

Lawrence R. Gordon

Psychology Research Methods I

Bystander Response to Arterial Bleeding Shotland & Heinold (1985) Research question? Methodological issues (relating to

participants’ rights, informed consent, deception, etc.)

Overall YEA/NAY? Revisions?

Southern Culture of Honor

Cohen, Nisbett, Bowdle, & Schwarz (1996) Research question? Methodological issues (relating to

participants’ rights, informed consent, deception, etc.)

Overall YEA/NAY? Revisions?

WHY REVISIT THIS NOW?

You are about to conduct research at UVM We want you to know more about the local

process You have learned more since the earlier

introduction that may provide a context Resources: UVM Human Subjects site:

– http:/www.uvm.edu/~reshmpg/test/irb-home.htm

– http://www.uvm.edu, then Research, Human Subjects

What Does the IRB Do?

Its chief function: Considers costs and benefits of the research

Asks, is the research question worth the use of human participants?

Because human participants do not need to participate in studies, their rights are the highest priority

Submitting Protocols In general --- wide range As part of a class

– Exempt from review– Expedited review– Full review: IRB meetings

Strong focus on Informed Consent– Lay summary– Consent Form– Often combined

Lay Summary No jargon! (hence “lay”) Elements:

• Title

• Invitation to participate

• Aims - hypothesis

• Background - WHY conducted

• Procedures - include time commitment

• Risks/Discomforts/Inconveniences

• Benefits - personal & societal

• Costs

• Many optional elements

Statement of Consent

Elements• Have read lay summary

• Understand procedure, risks, and benefits

• Participation voluntary; may withdraw any time

• Confidentiality to extent of the law

• Whom to contact if questions

• Signature

• Sometimes sign certifying a debriefing was given

Example of combined form -- Goodwin p. 50• For simple exempt study not terribly complicated

Issues or questions?

Yes? No? Then we’ll move on to some further ideas

in statistics that may be of help in understanding your analyses and output

Some ideas behind the statistics

Nature of “test statistics” (vs. descriptive)– e.g., t and F, so far….

Suppose the null hypothesis is true, what is the value of “Treatment”?

Suppose “Treatment”=0, what is the value of TestStat?

What happens as “Treatment” gets larger: to TestStat? to prob(TestStat|Null true) -- “p=”?

T estS ta tV ariab ility T rea tm en t E rror

V ariab ility E rror

( )

( )

Some ideas behind the stats (cont.)

What is this “df” thing?– E.g., , for n scores

– df = n-1 here, why? what’s it mean?

– Kinda “techie,” but if the mean, X-bar, is known, then only n-1 scores are “free to vary,” hence only n-1 “degrees of freedom” or “df”

– Example -- suppose you know the mean of 3 scores is 10, then if 2 are: the third must be:

12, 8 ?8,7 ?

13, 12 ?

V ar

SS

d f

X X

n

2

1

So, df in articles, etc.? Can be useful...

For independent groups means ---– t(28) means there were 30 scores, because for

this, df=(n1-1)+(n2-1)= n1+n2-2

For paired means (repeated measures)– t(28) means there were 29 pairs of matched

scores, df = n-1 pairs of related scores

Examples (blasts from the past)...

ANSWERS REVISITED“Having Fun” Example

Inferential Statistics

Independent Samples Test

-6.353 98 .000 -3.880Equal variancesassumed

Estimate of 10minute interval

t df Sig. (2-tailed)Mean

Difference

t-test for Equality of Means

Group Statistics

50 8.604 2.722

50 12.484 3.353

Experimental Conditions'More fun' (Captions)

'Less fun' (No Captions)

Estimate of 10minute interval

N Mean Std. Deviation

Repeated-measuresDefinitional Example “Family therapy for anorexia” (1994) SPSS -- standard analysis for paired-samples:

Paired Samples Statistics

83.229 17 5.017

90.494 17 8.475

Before therapy

After therapy

Pair1

Mean N Std. Deviation

Paired Samples Correlations

17 .538 .026Before therapy& After therapy

Pair1

N Correlation Sig.

Paired Samples Test

-7.265

7.157

-4.185

16

.001

Mean

Std. Deviation

Paired Differences

t

df

Sig. (2-tailed)

Before therapy- After therapy

Pair 1

So, df in articles, etc.? Can be useful…(continued)

For k independent groups means ---– “F(2,24)” means that there were

• 3 levels of the IV “Effect” df = k-1

• 24 “df for error” ”Error” df = 24 here

• 27 scores in all Effect df + Error df = N-1Source df

Effect 2Error 24

Total 26

• If equal size groups, how many Ps per group?

Example... “F(2,215)=5.314”

MEM 2002: Between-GroupsDescriptives: Total # Correct, both Trials

TOTAL

73 28.7945 5.65184 16.00 41.00

73 31.9041 5.13746 16.00 42.00

72 29.7361 6.82372 8.00 41.00

218 30.1468 6.02503 8.00 42.00

Non-specific

Imagery Instructions

Imagery Instructionsplus Picture

Total

N Mean Std. Deviation Minimum Maximum

ANOVA

TOTAL

371.070 2 185.535 5.314 .006

7506.233 215 34.913

7877.303 217

Between Groups

Within Groups

Total

Sum ofSquares df Mean Square F Sig.

So, df in articles, etc.? Can be useful... (continued) For k levels in repeated-meas ANOVA ---

– F(2,24) means that there were• 3 levels of the IV “Effect” df = k-1• 24 “df for error” ”Error” df = 24 here• Error = 2(n-1)=24, so n-1=12 13 Participants!• ?? scores in all Ss df + Effect df + Error df = N-1 Source

df Subjects n-1

Effect k-1

Error (n-1)(k-1)

Total nk-1 = N-1 scores

Example... “F(3,69)=5.60”

Mandel et al. (1995), from Handout last class: “Listening times to sound stimuli” “Across all 24 subjects {itemized values of 4 means}…an

[ANOVA] revealed …{means}were significantly different with a main effect of name category, F(3,69)=5.60, p=.0017.”

ANOVA Summary TableSource df SS MS F p

Subjects 23Listen Time 3 5.60

.0017 Error 69 Total 95

For “F(3,69), how many scores were there? 3+69=72 + 24 Ss = 96 (24 Ss x 4 scores each!)

EXAMPLE (SPSS -- Memory)…”F(2,434)=27.562”

MEM 2002: Within-Ss

Serial Position Curve

THIRDS

321

Me

an

# C

orre

ctly R

eca

lle

d

10.8

10.6

10.4

10.2

10.0

9.8

9.6

9.4

9.2

Tests of Within-Subjects Effects

Measure: MEASURE_1

176.113 2 88.057 27.562 .000

1386.554 434 3.195

SourceTHIRDS

Error(THIRDS)

Sum ofSquares df Mean Square F Sig.

Descriptive Statistics

10.7248 218

9.4633 218

9.9587 218

Total of 1st Third

Total of 2nd Third

Total of 3rd Third

Mean N

MEM 2001: Within-Ss2. THIRDS

Measure: MEASURE_1

10.465

8.985

9.355

THIRDS1

2

3

Mean

MEM2001: Mean correct by Thirds

Serial Position Curve

THIRDS

321

Me

an

# C

orre

ctly R

eca

lle

d

11.0

10.5

10.0

9.5

9.0

8.5

Tests of Within-Subjects Effects

Measure: MEASURE_1

237.293 2 118.647 29.390 .000

1606.707 398 4.037

SourceTHIRDS

Error(THIRDS)

Sum ofSquares df Mean Square F Sig.

N=200 Ps

Some ideas behind the Statistics…POWER

Recall the abstract definition of a test statistic:

We want to find effects if they’re there, and Power is the probability of doing that.

The larger the test statistic, the greater our chance of doing that.

Therefore, we want to maximize the numerator and minimize the denominator, but how?

T estS ta tV ariab ility T rea tm en t E rror

V ariab ility E rror

( )

( )

Influences on Power of the NHST … Pr(Reject null|Null false)

Level of significance used (); e.g. more power if .05 than if .01 (set a priori)

Size of the treatment effect: more power if larger effects (increase numerator)

Size of the sample: more power if N larger (decrease denominator by increasing df for error)

Experimental control and procedure (increase power by decreasing error variability in denominator)

Choice of design -- often within-Ss more powerful by reducing individual differences -- “error variance”

Review Goodwin pp. 136-141.

WRAPUP

Will go on to final major experimental design next two classes --- factorial designs and their interaction effects. Extremely important --- most frequently used designs!