You are on page 1of 4

Assignment 2

• Question 1 & 3 are covered in the tutorial.


• Question 2 and 4 are exercises to be completed on your own. You may submit
them to the TA if you want feedback (this has no impact on your final grades).
• If you don’t have much time, SKIP QUESTION 4.
• You can use the dataset adliking.sav if you want to practice factorial designs
with SPSS

Question 1
We investigate the impact of advertisement content on intention to purchase a chocolate truffle.
We alter the information content as low, moderate, and high, and observe consumers’
intentions.
The dataset (adinfo.sav) includes the following variables:
• Subject: id of the participant
• Advertisement: 1 if no information content, 2 if moderate, 3 if high
• PurchaseIntent: intention to purchase the product (ordinal variable between 1 and 7)
• Chocfan: General liking for chocolates
• Lastmeal: Minutes since last meal

A 3 by 1 ANOVA design is implemented.


a) Manually compute the within-groups Sum of Squares, the total Sum of Squares (use
excel) and thus the between-groups Sum of Squares. Compute the mean square error
by dividing the Sum of Squares by the corresponding degrees of freedom. Compute the
F-test.
Note: manual computations are not exam material. However, the exam material
includes knowing how to interpret ANOVA output in SPSS and the relation between
the items of the ANOVA table.
b) Carry out a one-way ANOVA test. Compare the results with parts a) and interpret the
resulting output.
c) Suppose that the estimated variance across either treatment group were larger (while
the total variance is the same). Will the within-groups Sum of Squares be larger or
smaller? Can you predict whether the null of “no treatment effect” will be rejected?
d) Plot the marginal means of intention to purchase. Perform contrasts and interpret the
resulting output.
e) In this setting, we are testing the difference between three treatments. Why would you
use a correction for multiple hypothesis testing? Implement it, and interpret the
resulting output.
Hint: https://towardsdatascience.com/an-overview-of-methods-to-address-the-multiple-
comparison-problem-310427b3ba92

f) Now carry out a one-way ANCOVA test, controlling for the impact of “chocfan” and
“Lastmeal”. Plot the marginal means of intention to purchase. Perform contrasts and
interpret the resulting output. Finally, use a correction for multiple hypothesis testing.

Question 2
Use the dataset music.sav for this question. Suppose that a manager of a clothing store chain
would like to find out whether or not the presence of music in the retail stores will result in
high sales. To do this, he selects 30 different stores which are similar in nature. 10 of the stores
does not have any music and this group also serves as the benchmark for the study. In 10 other
stores, he has a music in low volume. In the remaining stores, he has the same music in high
volume. Previous research has shown that consumers are likely to shop longer in stores with
low volume music than in stores with high volume music.

The stores used in the analysis have different surface areas and that previous research has
already shown that larger stores generate greater sales. In order to free the experiment with the
music from any potentially biasing effect of store size, the last aspect is included as a covariate
in the gathering and analysis of data.

Models
1. Perform an ANOVA (Analysis of Variance) to find out if sales is significantly different
between stores with use of high volume music, low volume music or no music. Next, using
contrasts, find out whether low and high volume music lead to better sales when compared to
stores with no music. Finally, using post-hoc tests, find out whether there is a difference in
sales between the low and high volume groups.
2. Perform an ANCOVA (Analysis of Covariance) to find out if sales is significantly different
between stores with use of high volume music, low volume music or no music controlling for
the covariate store size. Perform contrasts and post-hoc tests as above.

Questions
a) Are the sales of the three groups of stores (none, low volume music and high volume music)
different from each other?
b) Which of the groups have the highest sales?
c) Explain the differences in the findings between the two models (ANOVA versus ANCOVA).

Question 3
We will reconsider one of the studies from the article "Persuasion, interrupted: The effect of
momentary interruptions on message processing and persuasion" by Kupor and Tormala
(2015). The past week we have reconsidered study 1 from the article. There we saw that
momentary interruptions while exposed to an advertisement may increase willingness to pay
for a product.

Now we reconsider whether the impact of interruptions depend on a) the length of the
interruption – is it a short momentary interruption or a continuous distraction? - and b) the
content of the message - does the message require deep processing? We replicate the main
results of study 5, that uses a 2 by 2 factorial design to answer the two research questions.
Participants had to evaluate their attitude towards lowering the drinking age after having read
a short articles on the topic. Two different articles where used, one with a stronger argument
(requiring deeper processing) and one with a weaker argument. Participants were also
interrupted with either a short, momentary distraction, or a continuous one. For more details,
see the original article.

The dataset (KT2study5Qs3.xlsx) includes the following variables:


• Subject: id of the participant
• Attitude: attitude towards lowering the drinking age, where a higher number indicates a more favorable
attitude (unweighted average of various items, e.g. unfavorable-favorable, negative-positive, etc)
• Momentary interruption: 1 if short interruption, 0 otherwise (continuous distraction
• Weak argument: 1 if the message is simple, requiring little time to process it, 0 otherwise

1. How do attitudes differ across treatment? Choose a suitable technique, implement it and
interpret the resulting output. Motivate your answer.
2. Kupor and Tormala also measure behavioural intentions. Would it be a good idea to use an
ANCOVA design, controlling for behavioural intentions? Would you expect the results of
question 1 to change? Motivate your answer.

Question 4
Every year, the Economist Intelligence Unit (EIU) releases a new rank of global cities by
liveability. You can find the latest at: https://www.economist.com/graphic-
detail/2019/09/04/vienna-remains-the-worlds-most-liveable-city

In 2012, the EIU ran a competition for improving the EIU liveability index. The winner,
Filippo Lovato, proposed a liveability index that include spatial qualities of the cities. For
more info, read the corresponding press release:
https://www.economist.com/graphic-detail/2012/07/03/live-and-let-live
We use the data to study liveability across major world cities in 2012.

The dataset (EIUworldliveability.xlsx) includes the following variables:


• City: city of interest
• Region: the region where the city is located (1 = Western Europe, 2 = US & Canada, 3 = Asia, 4 =
Oceania, 5 = Eastern Europe, 6 = Latin America, 7 = Middle East, 8 = North Africa, 9 Sub-saharan
Africa)
• Rank – SpAdj Liveability index: the city score of the spatially adjusted liveability index
• Rank – EIU Liveability index: the city score of the Economist Intelligence Unit original liveability
index.
• Connectivity: the connectivity of the city, that is, the ease to reach the rest of the world (based on
nearby airports traffic)

Do not worry about the other variables

1. Use boxplots to plot the rank of EIU liveability of world cities across region. Also, discuss the
variation observed across each region.
2. Does liveability of cities change across macro-regions? Consider the same regions used in the
latest ranking (Western Europe, US & Canada, Asia & Australasia, Eastern Europe, Latin
America, Middle East & North Africa, Sub-Saharan Africa). Plot the marginal means across
macro region, perform contrast, and interpret the resulting output. Also implement a correction
for multiple hypothesis testing.
3. Using the information from question 2, compare the 2012 ranking with the current ranking. Are
there noticeable differences?
4. Now, use the spatially adjusted liveability index to consider whether the rank of world cities
changes across macro-regions. Are there noticeable differences?
5. Use a scatter plot to compare the EIU liveability index and the spatially adjusted liveability
index. Report the correlation: does the result in question 5 surprise you?
6. Answer again question 2, but now control for the impact of the variable “connectivity”

You might also like