Analysis > Multiple regression analysis

This report contains analysis of an online survey to copy-test three print advertisements for over-the-counter pain relief medications.  The online survey was distributed via e-mail and 126 usable responses were collected.  In this report, the results of the survey are subjected to multiple linear regression analyses for each brand.  The independent variables are ten Likert questions about attributes of each brand under investigation.  The dependent variable is the change in the respondent’s intention to purchase the brand after exposure to the three advertisements on a constant-sum scale.

Table 1: Multiple linear regression analysis of Advil, Bayer and Tylenol

Brand	Multiple correlation coefficient (R)	Coefficient of multiple determination (R²)	Standard error of the estimate (Se)	F
Advil	0.3	11.9%	2.4	1.55*
Bayer	0.3	7.3%	2.2	0.90
Tylenol	0.3	9.5%	2.5	1.21
				* p ≤ 0.15

Table 2: Advil brand attributes

Advil...	Unstandardized coefficient (b)	Standardized coefficient (β)	t
(Constant: 1.3)			0.69
...is a good brand	-1.6	-0.4	3.07*
...works fast	0.4	0.1	0.85
...prevents heart attacks	-0.3	-0.1	1.18
...is safer	-0.2	-0.0	0.43
...is trustworthy	0.2	0.1	0.64
...would be prescribed by my doctor	-0.0	-0.0	0.13
...is easy to use	-0.2	-0.1	0.53
...doesn’t have too many side effects	0.6	0.2	1.58*
...is made by a company that cares about my health	0.7	0.4	1.90*
...provides effective pain relief	-0.1	-0.0	0.11
			* p ≤ 0.15

In 85 samples out of 100 samples drawn from the same population as this sample, it would be expected that the ten Advil brand attitude items would account for a very small portion, about 11.9%, of the variance in the pre/post change scores for Advil.  The results of this sample can be projected to the population for this test.  However, only three of the brand attributes (Advil is a good brand, Advil doesn’t have too many side effects and Advil is made by a company that cares about my health) had a statistically significant impact; in 85 samples out of 100 samples drawn from the same population as this sample, it would be expected that these three coefficients would be about what they are here.

Table 3: Bayer brand attributes

Bayer...	Unstandardized coefficient (b)	Standardized coefficient (β)	t
(Constant: 0.8)			0.04
...is a good brand	1.1	0.3	2.00*
...works fast	-0.8	-0.2	1.57*
...prevents heart attacks	0.3	0.1	0.86
...is safer	0.4	0.1	1.01
...is trustworthy	0.1	0.0	0.19
...would be prescribed by my doctor	-0.3	-0.1	-0.95
...is easy to use	0.1	0.1	0.46
...doesn’t have too many side effects	-0.1	-0.0	0.24
...is made by a company that cares about my health	-0.2	-0.1	-0.47
...provides effective pain relief	-0.3	-0.1	0.63
			* p ≤ 0.15

In this sample, the brand attributes for Bayer accounted for a very small portion (7.3%) of the variance in pre/post change scores, but this is not statistically significant and cannot be projected to the population.  Only two of the brand attributes – Bayer is a good brand and Bayer works fast – had a statistically significant impact; in 85 out of 100 samples drawn from the same population as this sample, it would be expected that these two coefficients would be about what they are here.

Table 3: Tylenol brand attributes

Tylenol...	Unstandardized coefficient (b)	Standardized coefficient (β)	t
(Constant: -0.0)			0.01
...is a good brand	-0.5	-0.1	0.85
...works fast	0.1	0.0	0.34
...prevents heart attacks	0.0	0.0	0.08
...is safer	0.1	0.0	0.41
...is trustworthy	0.1	0.0	0.21
...would be prescribed by my doctor	0.1	0.0	0.38
...is easy to use	1.0	0.3	2.84*
...doesn’t have too many side effects	-0.7	-0.2	1.65*
...is made by a company that cares about my health	-0.2	-0.0	0.52
...provides effective pain relief	-0.4	-0.1	0.82
			* p ≤ 0.15

In this sample, the brand attributes for Tylenol accounted for a very small portion (9.5%) of the variance in pre/post change scores, but this is not statistically significant and cannot be projected to the population.  Only two of the brand attributes – Tylenol is easy to use and Tylenol doesn’t have too many side effects – had a statistically significant impact; in 85 out of 100 samples drawn from the same population as this sample, it would be expected that these two coefficients would be about what they are here.