The Influence of Suggestion on Subjective Preferences By Sean Oh, Joshua Marcuse, and David Atterbury Math 5: Chance.

The Influence of Suggestion on Subjective PreferencesBy Sean Oh, Joshua Marcuse,

and David Atterbury

Math 5: Chance

Goal

Hypothesis: Social conformity is an essential trait of human nature.

Subjective preferences may be susceptible to suggestion when this trait is activated.

The goal of the experiment was to show that subjective preferences can be swayed by suggestion.

Experimental Design

Two Pictures of female models: A and B. Three Treatments: A, B, and Control. In each treatment we ask males to state

which model they think is more beautiful. However, in Treatments A and B we tried to

influence the respondent’s preference with a suggestion to see if it affected his answer.

Picture A

Picture B

Null Hypothesis

Respondents in Treatment A and Treatment B are equally likely to prefer model A or model B as they did in Treatment Control.

Alternate Hypothesis

Respondents in Treatment A will tend to prefer model A and respondents in Treatment B will tend to prefer model B compared to the Treatment Control.

How did we try to influence them?Our script for Treatments A and B said:

“Hello. I am conducting a psychology experiment. Would you please look at these two pictures. In our recent study, a majority of people stated that the woman in Picture A [or B] is more beautiful. Do agree or disagree that the woman in Picture A [or B] is more beautiful?”

Treatment Control

For the Control we tried to establish a baseline against which we could compare the results of Treatments A and B.

We did not make any suggestion to attempt to influence the respondent.

We hoped to get as close to 50% as possible for Pictures A and B.

Treatment Control Script

Our script for Treatment Control said:

“Hello. I am conducting a psychology experiment. Would you please look at these two pictures and tell me if you think the woman in Picture A is more beautiful, or do you think the woman in Picture B is more beautiful?”

How we collected the data We interviewed 135 people for the experiment. Each treatment contained 45 respondents. All respondents were RANDOMLY selected. We collected data in Thayer, Collis and Novack, during the

morning, afternoon, and evening. 15 respondents were interviewed at each location. Then we

aggregated the data so all three Treatments included data taken from all three locations during all three times of day.

We used three interviewers to administer the question from the script.

Each respondent was interviewed separately, and additional precautions were taken to avoid any external influence on the respondent during the experiment.

Why we excluded women

We wanted to include men and women in our study, but…

When we asked women in pre-test whether they preferred Model A or Model B, we got a surprising result…

Massive bias among femalesPre-Tests Results

02468

10121416

women men

Resp

onde

nt's

Pre

fere

nces

Model in Picture A

Model in Picture B

The Math…

Parameter

Treatment Control gauged the parameter of preference for model A and model B.

N = 45

Preference for model A = 24/45 = .533 Preference for model B = 21/45 = .467

Significance Level and Critical Region

Significance Level = 3.67%

Critical Region for Treatment A: PA ≥ 30 people

Critical Region for Treatment B: PB ≥ 27 people

What does that mean?

If 30 or more respondents choose model A in Treatment A and if 27 or more respondents choose model B in Treatment B, we can say with over 95% certainty that subjective preferences were influenced by our comments.

Power

We chose .7 as a power. We believed that 70% of respondents would choose model A in Treatment A and 70% of respondents would choose model B in Treatment B.

Using this power, we found that there would be a 31.21% chance of a Type II error in Treatment A and a 7.21% chance of a Type II error in Treatment B.

What does this mean?

According to our power, if 70% of people truly preferred model A in Treatment A, we have about a 31% chance of not reaching the critical value and thus incorrectly concluding that respondents were not influenced.

Same for model B in Treatment B, except this is only a 7% chance.

Results

Treatment A: Preference for model A: 33/45 = .733 Preference for model B: 12/45 = .267

Treatment B: Preference for model A: 30/45 = .667 Preference for model B: 15/45 = .333

Preferences for Model A and B

05

101520253035

TreatmentA

TreatmentB

TreatmentControl

AB

Analysis of Results

Our results were certainly surprising. Just looking at the numbers, we can say with

96% certainty that people were influenced in Treatment A AND we can say with 93% certainty that people were NOT influenced in Treatment B.

In conclusion, the data does not support our hypothesis at all.

Possible explanations

Sample size was too small to indicate the subtlety of our hypothesis.

Treatment Control misrepresented the population. People actually preferred model A to model B at a 2:1 ratio, but we only got a 1:1 ratio by chance.

More Possible Explanations Bias in test administration

Suggestion influenced the respondents, but not in the way we predicted