Exercise 3 Answer Key

Question 2
We reconsider study 5 from the article "Persuasion, interrupted: The effect of momentary
interruptions on message processing and persuasion" by Kupor and Tormala (2015). We have
already analyzed this study in the past week using ANOVA. This week we use linear regression.
Download the dataset used in week 2: KT2study5Qs3.xlsx

a) Create a new dummy variable for the interaction effect between “Momentary
interruption” and “Weak argument”. Then, estimate the following model:
𝐀𝐭𝐭𝐢𝐭𝐮𝐝𝐞 = 𝜷𝟎+𝜷𝟏𝐌𝐨𝐦𝐞𝐧𝐭𝐚𝐫𝐲 +𝜷𝟐𝐖𝐞𝐚𝐤 +𝜷𝟑𝐌𝐨𝐦𝐞𝐧𝐭𝐚𝐫𝐲∗𝐖𝐞𝐚𝐤

b) Interpret all coefficients: which effects are significant? In particular:

o What is the expected attitude of a participant who is momentarily interrupted
when the argument is weak?
o What is the expected attitude of a participant who is continuously interrupted
when the argument is strong?

Assuming a significance level of .05:

• Momentary interruption has a significant effect on sales when the argument is strong (weak
arguments = 0, p-value = 0.000 < .05)
• Weak arguments does not have a significant effect on sales when the interruption is long
(momentary interruption = 0, p-value = 0.138 > .05)
• The interaction between momentary interruption and weak arguments has a significant
impact of sales (p-value = 0.038 < .05)

The expected attitude of a participant who is momentarily interrupted when the argument is weak =
4.304 + 1.396×1 + 0.528×1 – 1.046×1×1= 5.18
The expected attitude of a participant who is continuously interrupted when the argument is strong
=
4.304 + 1.396×0 + 0.528×0 + −1.046×0×0 = 4.304

c) Compare the results with the ANOVA output you obtained in exercise from the past week.
Comment on similarity and differences.
There’s more than one way to answer this question. The ANOVA output is given by:

We may compare the regression output with that shown in the histogram in the previous week.
Notice that expected attitude of a participant who is momentarily interrupted (red color) when the
argument is weak (right bars) is approximately 5.18 in the picture, which is the number we obtained
in part b. Alternatively, the expected attitude of a participant who is continuously interrupted (blue
color) when the argument is strong (left bars) is approximately 4.304, as obtained earlier.

Notice that the ANOVA output however does not strictly correspond to the output of the linear
regression. This is because ANOVA is testing a main effect of each manipulation (regardless of
whether the other manipulation is present, or not). Linear regression is instead testing the impact of
either manipulation when the other manipulation is not present. Hence, the significance levels in the
ANOVA table do not correspond to those obtained by linear regression for the respective factors
(with the exception of the interaction term).
Tests of Between-Subjects Effects
Dependent Variable: attitudes
Type III
Sum of Mean Partial Eta
Source Squares df Square F Sig. Squared
Corrected Model 51.784a 3 17.261 5.494 .001 .078
Intercept 5009.004 1 5009.004 1594.33 .000 .891
9
momentaryinterrupti 38.106 1 38.106 12.129 .001 .058
on
weakarguments .001 1 .001 .000 .984 .000
momentaryinterrupti 13.676 1 13.676 4.353 .038 .022
on * weakarguments

Error 615.782 196 3.142

Total 5676.570 200
Corrected Total 667.566 199
a. R Squared = .078 (Adjusted R Squared = .063)

Question 3
Output of linear regression analysis

Model Summary

Std. Error of the

Model R R Square Adjusted R Square Estimate

1 .733a .537 .528 .32849

a. Predictors: (Constant), CompFeature, CompDisp, lnOwnPrice, OwnDisplay, lnCompPrice,

OwnFeature

ANOVAa

Model Sum of Squares df Mean Square F Sig.

1 Regression 36.568 6 6.095 56.482 .000b

Residual 31.508 292 .108

Total 68.076 298

a. Dependent Variable: lnSales

b. Predictors: (Constant), CompFeature, CompDisp, lnOwnPrice, OwnDisplay, lnCompPrice, OwnFeature

Coefficientsa

Standardized

Unstandardized Coefficients Coefficients

Model B Std. Error Beta t Sig.

1 (Constant) 7.372 .073 101.461 .000

lnOwnPrice -.794 .198 -.186 -4.010 .000

lnCompPrice 2.195 .404 .252 5.432 .000

OwnDisplay .547 .048 .558 11.408 .000

OwnFeature .127 .047 .133 2.682 .008

CompDisp -.391 .045 -.406 -8.733 .000

CompFeature .070 .043 .073 1.642 .102

a. Dependent Variable: lnSales

1. What is the own price elasticity of demand for C-Cola? In other words, what % will the sales of
C-Cola change if the price of C-Cola is increased by 1%?

%∆𝑆𝑎𝑙𝑒𝑠 (∆𝑆𝑎𝑙𝑒𝑠/𝑆𝑎𝑙𝑒𝑠) 𝑑(ln(𝑆𝑎𝑙𝑒𝑠))

Price elasticity = = =
%∆𝑃𝑟𝑖𝑐𝑒 (∆𝑃𝑟𝑖𝑐𝑒/𝑂𝑤𝑛𝑃𝑟𝑖𝑐𝑒) 𝑑(ln(𝑃𝑟𝑖𝑐𝑒))

Since the regression model is given as

ln(𝑆𝑎𝑙𝑒𝑠𝑡 ) = 𝛽0 + 𝛽1 ln(𝑂𝑤𝑛𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽2 ln(𝐶𝑜𝑚𝑝𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽3 𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 + 𝛽4 𝑂𝑤𝑛𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡
+ 𝛽5 𝐶𝑜𝑚𝑝𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 + 𝛽6 𝐶𝑜𝑚𝑝𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡 + 𝜀𝑡

𝑑(ln(𝑆𝑎𝑙𝑒𝑠))
We can get own price elasticity 𝛽1 = 𝑑(ln(𝑂𝑤𝑛𝑃𝑟𝑖𝑐𝑒)) = 𝛽1 (ceteris paribus)

If the price of C-Cola increases by 1%, the sales of C-Cola will decrease by 0.794%, ceteris paribus.

2. What is the competitive price elasticity of demand for C-Cola? In other words, what % will the
sales of C-Cola change if the average price of competitors is increased by 1%?

Similar to the answer of Q1: If the average price of competitors increase by 1%, the sales of C-Cola will
increase by 2.195%, ceteris paribus.

3. What will be the sales (in liters) of C-Cola when it is displayed in end-of-aisle or at checkout
counters if the sales when C-Cola is not displayed is 1000 liters?

When 𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 0,

ln(𝑆𝑎𝑙𝑒𝑠𝑡 ) |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 0)
= 𝛽0 + 𝛽1 ln(𝑂𝑤𝑛𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽2 ln(𝐶𝑜𝑚𝑝𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽4 𝑂𝑤𝑛𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡
+ 𝛽5 𝐶𝑜𝑚𝑝𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 + 𝛽6 𝐶𝑜𝑚𝑝𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡

When 𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 1,

ln(𝑆𝑎𝑙𝑒𝑠𝑡 ) |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 1)
= 𝛽0 + 𝛽1 ln(𝑂𝑤𝑛𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽2 ln(𝐶𝑜𝑚𝑝𝑃𝑟𝑖𝑐𝑒𝑡 ) + 𝛽3 + 𝛽4 𝑂𝑤𝑛𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡
+ 𝛽5 𝐶𝑜𝑚𝑝𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 + 𝛽6 𝐶𝑜𝑚𝑝𝐹𝑒𝑎𝑡𝑢𝑟𝑒𝑡

Difference in sales = ln(𝑆𝑎𝑙𝑒𝑠𝑡 ) |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 1) − ln(𝑆𝑎𝑙𝑒𝑠𝑡 )|(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 0) = 𝛽3

𝑆𝑎𝑙𝑒𝑠 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦 =1) 𝑆𝑎𝑙𝑒𝑠 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦 =1)

⇒ ln (𝑆𝑎𝑙𝑒𝑠𝑡 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 =0)) = 𝛽3 ⇒ 𝑆𝑎𝑙𝑒𝑠𝑡 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 =0) = 𝑒 𝛽3
𝑡 𝑡 𝑡 𝑡

⇒ 𝑆𝑎𝑙𝑒𝑠𝑡 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 1) = 𝑒 𝛽3 × 𝑆𝑎𝑙𝑒𝑠𝑡 |(𝑂𝑤𝑛𝐷𝑖𝑠𝑝𝑙𝑎𝑦𝑡 = 0) = 1000𝑒 𝛽3

Therefore, the sales (in liters) of C-Cola when it is displayed in end-of-aisle or at checkout counters is
1000𝑒 0.547 = 1728.061 𝑙𝑖𝑡𝑒𝑟𝑠
4. What will be the sales (in liters) of C-Cola when the competitor’s product is displayed in end-of-
aisle or at checkout counters if the sales of C-Cola without competitor’s product being displayed
is 1000 liters?

The sales (in liters) of C-Cola when the competitor’s product is displayed in end-of-aisle or at checkout
counters is 1000𝑒 −0.391 = 676.3802 𝑙𝑖𝑡𝑒𝑟𝑠.

5. What will be the sales (in liters) of C-Cola when it is featured in the flyer if the sales of C-Cola
without it being featured in the flyer is 1000 liters?

The sales (in liters) of C-Cola when it is featured in the flyer is 1000𝑒 0.127 = 1135.417 𝑙𝑖𝑡𝑒𝑟𝑠

6. What will be the sales (in liters) of C-Cola when the competitor’s product is featured in the flyer
if the sales of C-Cola without competitor’s product being featured is 1000 liters?

The sales (in liters) of C-Cola when the competitor’s product is featured in the flyer is 1000𝑒 0 =
1000 𝑙𝑖𝑡𝑒𝑟𝑠. If a parameter is insignificant, it is not significantly different from zero.

7. Among the independent variables, which of them has the highest impact on C-Cola’s sales, and
which of them has the lowest but significant impact on C-Cola’s sales?

Based on the standardized beta, Own Display has the highest and Own Feature has the lowest
significant impact on C-Cola’s sales.

Question 4

a) A = 566; B = 3; C = 296; D = 189

b) Picture 1 corresponds to model 1, Picture 2 corresponds to model 3, and Picture 3 corresponds
to model 2.
Picture 1 shows a pretty good model fit, this corresponds to the significant F test in model 1.
Picture 2 shows a worse model fit, this corresponds to the insignificant F test in model 3.
Picture 3 shows that main effects exist for both X0 and X1, but the interaction seems have no
influence on attitude. This corresponds to model 2 where color significantly decreases attitude,
while females have a higher attitude than males. However, the interaction of female and color
did not have a significant coefficient.
c) Log(attitude) = 0,5 + 0,5 = 1 thus, attitude = 𝑒 1= 2,72
d) Log(attitude_male) = 𝛽0 + 𝛽1 *0 + 𝛽2 *1 + 𝛽3 * 1 * 0 = 𝛽0 + 𝛽2
Log(attitude_female) = 𝛽0 + 𝛽1 *1 + 𝛽2 *1 + 𝛽3 * 1 * 1 = 𝛽0 + 𝛽1 + 𝛽2 + 𝛽3
Log(attitude_female) – log(attitude_male) = 𝛽1 + 𝛽3
attitudefemale
= 𝑒 (𝛽1 + 𝛽3 )
attitudemale
attitudefemale
− 1 = 𝑒 (𝛽1 + 𝛽3 ) − 1
attitudemale
attitude_female – attitude_male
attitude_male
= 𝑒 (𝛽1 + 𝛽3 ) − 1
Thus the expected percentage change in attitude when the advertisement is shown to females instead of
males is exp(0,5 - 0,1) – 1= 0,4918 = 49.18%
e) Female has the highest positive impact on attitude (𝛽1 = 0,9)