Tequila Consumer Preference Survey

.

Enter

Table of Contents

Executive Summary

Please click here for the Introduction page.

Introduction

Please click here for the Methodology section.

Methodology

Structure of the Questionnaire

The consumer questionnaire was designed using Dreamweaver and analyzes consumer preference of three brands of tequila.  The layout was designed specifically for online data collection, with a survey title of "Consumer Preference Study" and it intentionally omitted identifying brands or products. The survey was constructed to seem as if there were multiple pages. Due to data collection concerns, however, it was actually one page. The questionnaire consisted of 12 sections, each section containing three to ten questions. Below you will find detailed descriptions of each section.

Title Page
The introductory page was the first page that the respondents saw upon clicking on the hyperlink in their email. It was designed to motivate participants and assure them of complete confidentiality of the survey.

Section I
The first section had three questions which asked respondents to type in last brand of beer, wine, and tequila they purchased. The purpose of this question is to determine which brands are "top of mind" to the respondent and to which brands the respondent feel most loyal. At this point, the respondent the survey didn't reveal the focus of the study. The goal was to minimize biases when collecting this data.

Section II
The questions in this section asked the respondents to divide 10 points among three different brands of wine, three different brands of beer, and three different brands of tequila, based on how likely they were to purchase each distinct product. The values from this scale are eventually used to analyze pre- and post-ad exposure changes. The respondents were then exposed to three print ads. The respondents are told to view the ads only for approximately 30 seconds. This replicated the viewing environment as if respondents were flipping through a magazine, much like the one they were taken from. 

Section III
After viewing the ads, the respondents were asked to indicate purchase intention for each brand of tequila in the same manner as described above. This "post-exposure" question allows for comparisons of pre- and post-exposure scores in order to obtain a change score for each brand of tequila.

Section IV
This section asked respondents to indicate whether or not viewing the ad changed their opinion of the brand. If there was no difference, the respondent skipped ahead to Section V. If the respondent marked yes, he or she is then asked which brand was affected, whether the impact was favorable or unfavorable, and to note any other comments about the difference of opinion.

Section V
This section established brand attitudes of respondents. There were ten statements about each brand of tequila followed by a Likert scale. They had options to check “strongly agree,” “agree,” “neither agree nor disagree,” “disagree,” or “strongly disagree.” Each scale point was coded as a numeric value for later statistical analysis. A value of 5 was given to “strongly agree,” and 1 to “strongly disagree." On positively worded phrases and a value of 1 is given to “strongly agree,” and 5 to “strongly disagree”on negatively worded phrases. The statements were as follows:

“Patron is a good tequila.
I prefer Patron to any other brand of tequila.
I would never drink Patron tequila.
Patron has a smoother, richer taste than other brands of tequila.
Patron is sexy.
Patron is too expensive.
Patron has a higher alcohol content than other tequilas.
I would recommend Patron tequila to others.
I would enjoy drinking Patron.
Patron is distilled from higher quality agave plants than other tequilas.”

Sections VI and VII
These two sections were identical to Section V, except that the statements used different brands. Section VI was about 1800 and Section VII about Jose Cuervo.

Section VIII
This section asked the respondents to check any statements they agreed with concerning three ads. The responses are numerically coded for quantitative analysis. The respondent is asked to respond to the following impressions: for each brand, was the advertising message about the product: Confusing, Important, Informative, Persuasive, Memorable, Unique, Interesting, Boring, Silly, Stereotypical

Sections IX, X and XI
Next, the respondent is asked to check a box if they agreed with the statement beside it about the brand’s ad.  These questions pertain to specifics about the ads' execution and are as follows:

“Quickly forget the ad.
Learn something new.
Find it memorable.
Discover an interest.
Want to see it again.
Find the ad annoying.
Feel disappointed.
Plan on buying the brand.
Think differently about the brand.
Think it would appeal to men.”

Final Section
 The purpose of this section is to examine if any demographic information is related to how the respondent reacts to the ad.

top

Design

This study is a survey. There is no control group to compare with this group of respondents. Comparison is made instead between before and after advertising exposure using the same sample. By subtracting pre-exposure purchase intention from post-exposure purchase intention, the constant sum scale measures the effects of adverting.

top

Sampling Description

Due to time and monetary constraints, this study opted for a non-random, convenience sample. It was decided that a sample size of at approximately 60 would be large enough to perform statistical analysis. This sample size also allows for enough representation so that if statistical significance is found, projection can be made in 85 or more samples of 100 samples within the same population as these 99 were drawn we would find results of the same magnitude.

An email was sent out to personal and professional acquaintances of the researcher. The email contained the link to the online survey asking individuals to complete the survey and to forward on to any other individuals who may be interested in participation. The respondents agreed to participate in the study by clicking the "Submit" button.

top

Data Collection Process

53 respondents successfully filled out this survey. The survey was designed in Dreamweaver and was connected to an MS Access database (.mdb) using a Coldfusion file (.cfm) so that the responses would be captured in the database. The responses were later transferred into SPSS (a statistical software package for the social sciences) for analysis. Due to this method of online data collection, error was minimized as the respondents' exact answers were transferred directly into the database and no manual coding was needed.

A web page was constructed to appear as if there were multiple pages in the questionnaire. The questionnaire had detailed instructions to guide respondents through to completion. The questionnaire employed text fields to gather written comments and drop-down menus or radio buttons, so respondents could select the appropriate answers as easily and efficiently as possible. When the questionnaire was created, each answer was given a field name and a coded number (if applicable) so that the numbered data could be quantitatively used in a statistical program for analysis.

top

Please click here for the Analysis section.

Analysis

An online consumer preference survey was conducted in September 2006, comparing three different brands of tequila. A total of 61 consumer responses were gathered, 53 of which were valid. The purpose of statistically analyzing this particular data set is to determine how three different print advertisements of three different brands affect consumer preference, perception, and likeability of the following brands: Patron, 1800, and Jose Cuervo. Here are the results of the statistical analysis.

Basic Statistics

1). Correlated (paired) t-test comparing brand index score for brands A, B, and C.

t- ratio

 t is reported in absolute value

t=0.016 = 0.008*
2
*p=< 0.15 alpha level

t=0.001 = 0.0005*
2
*p=<0.15 alpha level

t=0.564 = 0.282
2
*p=<0.15 alpha level

Jose Cuervo has the highest brand index score at 30.1 followed closely by Patron at 29.6 and 1800 at 27.9 respectively.  The former two are close to their brand index mean, which indicates parity among those two brands within the sample size, but respondents on average felt that 1800 ranked lower.  The negligible difference between Patron and Jose Cuervo means that in the general population one could not accurately hypothesize which brand the general population prefers between them, but that one could consistently expect a preference for them over 1800. This fact is reflected by the low t-ratios of Jose Cuervo to Patron.

2). Between groups t-test pre to post exposure, for only Brand A, Patron.  Equal variance not assumed, reported to two decimal points and in absolute value.

The brand index and ad index score for those respondents that were up-movers, pre to post exposure was higher for Patron relative to those who moved down.  This result is logical, and means that in 85 or more samples out of every 100 samples drawn from the same population as this sample, the mean brand index score for brand A, Patron, for the up and down movers would be about what it is in this sample. 

3). Chi-squared significance inference test for Brand A, Patron.  The median for Patron is 31.  The median is the 50th percentile and hence is a better indication of normal distribution than the mean.

Chi-Squared Test for Patron

Respondents that lowered their brand score for a given brand post advertisement exposure tended to have a brand index score over the median for that particular brand.  This result is surprising and does not allow for a definitive conclusion about why this is the case. 

4). 1800 median is 30.  Jose Cuervo median is 32. 

5). Brand A, Patron – Brand B, 1800
No table necessary. 

24 out of 53 valid responses have a higher Brand A, Patron brand index score than Brand B, 1800 brand index score.  This is valuable information as we try to understand what results are indicative of the advertisement and what results are indicative of prejudgment.

6). Simple Correlation coefficient between Brand A, Patron Index Scores and Brand B, 1800 Index Scores, it is a descriptive statistic – only one decimal point needed.

The correlation coefficient is a value between 1 and negative one.  The value from the data comparing Patron and 1800 brand index scores shows the correlation to be .8 which is a relatively strong correlation of statistical significance.  From this result we can conclude that within the general public there is a slight tendency for people to prefer Patron to 1800.  

Correlations between Brand Index Scores of Brands A & B

7). Basic statistical analysis with a subset of data, using only respondents who purchase alcohol three or more times a month.  Finding the correlation coefficient between Brand A, Patron and Brand B, 1800 among high frequency buyers, only 33 of the original 53 respondents answers will be considered.  The correlation is slightly higher here, at .9 instead of .8 as was the case using all the data, which is a logical that the significance would increase among heavy users of the product.  

Correlation coefficient:

top

Please click here for Multiple Regression.

Back to Table of Contents

Multiple Regression

Patron Table A                                                            

Patron Table B

*p = 0.15

In table A, R is the coefficient of multiple correlations, or the slope in the y = a+ bx line and is 0.6. This positive number means the line slopes upward, suggesting that as the people in the sample increased their opinion after viewing the ad for Patron, they also raised their opinion of the brand itself. Through R, the percentage of people whose Likerts scores can be used to predict a higher ad index score for the brand is at a percentage of 36% with a standard error of 1.7.

R-squared is the coefficient of multiple determination and can also be described as the explained variance divided by the total variance. The higher R-squared is, the less the unexplained variance. However, in the above sample, R-squared is low and therefore implies that the unexplained variance is large. Since standard error is the square-root of unexplained variance, the low R-squared in this sample also suggests a large standard error for the estimate.

The F-ratio tests the significance of R. As an inferential piece of data, it shows whether or not the results of the above cross-sectional sample can be reflected onto the population. However, in Table A, the F-ratio of 2.35 is not significant, and therefore the results cannot be projected onto the population.

In Table B, B is the y-intercept in the line y = a+ bx and is therefore the point in which the line intercepts the y-axis. In the above sample, the constant of B is -0.3 with a standard error of 0.9. Being positive, it is directly related to the regression coefficient (R).

The constant for B and its subsequent values comprise the constant term and the unstandardized regression coefficients for the equation. Standardized regression coefficients are depicted by Beta. Beta proves the relative importance of the independent variables such as the Likert items in relation to the variance of the dependent variable (or the change score). Dividing the biggest absolute value of Beta by 2, any number in the column that is 0.1 or bigger is considered important, and is therefore bolded.  The t-ratio, inferential like the F-ratio, proves whether or not the specific line would be in the same place if the dots for the entire population were present.

The Likert item with the lowest significance in the coefficients table (less than 0.05) is the one most correlated to a higher ad index score.  In Patron’s case, this is the Likert question asking for a ranking of the belief that Patron uses higher quality agave plants to make their brand of tequila.  The equation applying this would be the qualp Likert score multiplied X 1.0 – 0.9 = reasonable estimate of the dependant variable or ad index score for the brand.  

Patron Table C

1800 Regression

1800 Table A                                                                

1800 Table B

*p = 0.15

In Table A, R is set at 0.5. Since it is positive, this means the line slopes upward, suggesting that as the people in the sample increased their opinion after viewing the ad for 1800, they also raised their opinion of the brand itself. Through R, the percent of how much people like the ad can be explained by how much they like the brand. For 1800, the percentage is at 27%, with a standard error also at 1.7.

R-squared in this sample is relatively low, implying that the unexplained variance is large. Since standard error is the square-root of unexplained variance, the low R-squared in this sample also suggests a large standard error for the estimate.

In Table B, the constant of B is -0.1 with a standard error of 0.9. Being negative, it is inversely related to the regression coefficient. Beta proves the relative importance of the independent variables such as the Likert items in relation to the variance of the dependent variable (or the change score). Dividing the biggest absolute value of Beta by 2, any number in the column that is 0.2 or bigger is considered important, and is therefore bolded.  The t-ratio, inferential like the F-ratio, proves whether or not the specific line would be in the same place if the dots for the entire population were present.

The Likert item with the lowest significance in the coefficients table is the one most correlated to a higher ad index score.  In 1800’s case, this is the Likert question asking for a ranking of the belief that 1800 is stronger that other brands of tequila.  The equation applying this would be the stronge Likert score multiplied X 0.6 – 0.7 = reasonable estimate of the dependant variable or ad index score for the brand.  

1800 Table C

Jose Cuervo Regression

Jose Cuervo Table A                                          

Jose Cuervo Table B

 *p = 0.15         

In Table A, R is set at 0.5. This positive number means the line slopes upward, suggesting that as the people in the sample increased their opinion after viewing the ad for Jose Cuervo, they also slightly raised their opinion of the brand itself. Through R, the percent of how much people like the ad can be explained by how much they like the brand, standing for Jose at 20% with a standard error of 2.2. In the above sample, R-squared is medium and therefore implies that the unexplained variance is of relatively large size. Since standard error is the square-root of unexplained variance, the moderately low R-squared in this sample also suggests a somewhat large standard error for the estimate.

In Table B, the constant of B is 0.04 with a standard error of 1.2. Being positive, it is directly related to the regression coefficient. Beta proves the relative importance of the independent variables, the Likert items, in relation to the variance of the dependent variable. Dividing the biggest absolute value of Beta by 2, any number in the column that is 0.2 or bigger is considered important, and is therefore bolded. The t-ratio, inferential like the F-ratio, proves whether or not the specific line would be in the same place if the dots for the entire population were present.

Jose Cuervo Table C

top

Please click here for Discriminant Analysis.

Back to Table of Contents 

Discriminant Analysis

Discriminant analysis was run on the brand with the most movers, which in this case it was Patron tequila.  Very few of the respondents to the survey actually changed their opinion about the brand after seeing the ads.  Due to this, I recoded the non-movers to join the category of down movers, concluding that apathy should have a negative connotation.  The results for this analysis were obtained via an online consumer preference survey. Sixty-one respondents participated in the study, but only fifty-three responses were valid for all of the needed data for this analysis.

These two groups of up and down movers were used as the categorical dependent variable and the brand attributes (10 Likert items) were used as the independent variables. This analysis was conducted to determine whether or not we can accurately predict, random chance accuracy aside, the divide between these groups through their rating of brand attributes (Likert) responses.

Group Statistics

When comparing mean scores of brand attributes between those whose scores increased or decreased, there are no items that have dramatically different mean scores, thus there seems to not be much to discriminate. 

Wilks' Lambda

The significance value of 0.47 meets the required significance level of at least 0.15.  These results can be projected to the population at large from this sample.

Standardized Canonical Discriminant Function Coefficients

The largest value, taking the absolute value, is the first Likert item, which asks the respondent to score if they feel the ad for Patron is confusing. 

1.53/2 = 0.42 > 0.4 which shows a slight difference between the up and down movers.

Unstandardized Canonical Discriminant Function Coefficients

Functions at Group Centroids          

The group centroids are significant at 0.9 for those whose scores increased and -0.2 for those whose scores decreased, due to the significance found for Wilks' Lambda at 0.83. The Chi-squared, which indicates significance of Wilks’ Lambda, is significant and shows relationship between the respondents whose scores increased or decreased and the brand attributes.

Prior Probabilities for Groups

Classification Results(a)

a.83% of original grouped cases correctly classified.
t0 = t = (0.83 – 0.5) / √ [(0.83*0.17/53) + (0.5*0.5/53)]
= 0.33/√(0.0027 + 0.0047)
= 0.33/√0.0074
 = .33/.086       =          3.8
            T0        >>        tc         =          1.04      
As t0 is much greater than tc and the p-value is less than .001, the result for is highly significant and classification by chance alone is rejected.

Conclusions

top

Please click here for ANOVA/MANOVA.

Back to Table of Contents

ANOVA/MANOVA

Analysis of variance between groups tests a null hypothesis that several group means are equal in the population, by comparing the sample variance estimated from the group means to that estimated within groups. Using the brand attribute “This is a good brand of tequila,” as a dependent variable, and the income and intent to buy this brand as two independent variables, the two-way factorial ANOVA test attempted to determine their relationship. It will show which income levels of respondents had the best response.

When using a factorial design, we can draw conclusions regarding effects of the independent variables separately, as well as the combined effect of the independent variables. The two-way MANOVA took all ten brand attributes or Likert items as dependent variables for the same test. The analysis used 53 responses obtained through an online survey for the Jose Cuervo brand of tequila.

ANOVA Table

*p<= 0.15
The F ratios are significant, therefore we can conclude that in 85 or more samples out of every 100 drawn from the same population as this sample of 53, we would expect to find similar means scores and variances of the same magnitude.  From this analysis we can conclude that people who think this is a good brand of tequila are more likely to make under $40,000.

MANOVA
This procedure tests any of a wide variety of null hypothesis about the effect of other variables on the mean value of several correlated dependent variables.

*p<= 0.15
The mean scores of the relationships between the ten brand attributes, change score movement and those who intend to buy the brand are similar in value with fairly large deviations. The F ratios are both significant; therefore in 85 or more samples out of every 100 drawn from the same population as this sample of 53, we would expect to find similar means scores and variances of the same magnitude as in this study.

top

Please click here for Factor Analysis.

Back to Table of Contents

Factor Analysis

A factor analysis was performed for each brand of tequila to place related brand attributes into independent groups, or factors, for the calculation of a Brand Attitude Score. The variables used were 10 Likert items. A paired t-test was then performed to find potential statistical significance for the differences of means between brand attitude scores.

Patron Communalities

Extraction Method: Principal Component Analysis.

Patron Component Matrix

 For Patron 74 % of the variance can be explained by two factors (26 % is unexplained or lost variance). These two factors were the only to have Eigenvalues greater than one. The Factor Matrix was used in this analysis because it contained only one ambiguous row. The good item “this is a good brand of tequila” loaded on factor I. However, no other Likert items loaded on this factor. Therefore, only one item was used to compute a Brand Attitude Score of 3.02 for Patron.

1800 Communalities

Extraction Method: Principal Component Analysis.

1800 Component Matrix

For 1800, 76.7% of the variance can be explained by two factors. 23.3 % is unexplained or lost variance. Only one factors had an Eigenvalues greater than 1.0, so the matrix could not be rotated.  These 10 items were used to compute a Brand Attitude Score of 2.8 for 1800.

Jose Cuervo Communalities

Extraction Method: Principal Component Analysis.

Jose Cuervo Component Matrix

For Jose Cuervo 80.4% of the variance can be explained by two factors, and approximately 20 % is unexplained or lost variance. These two factors were the only ones that had Eigenvalues greater than 1.0. The Factor Matrix was used since again there was only one ambiguous rows. The good item “Jose Cuervo is a good tequila ” loaded on factor I and so did almost all of the rest of the Likert items, except for one. These 10 items were used to compute a Brand Attitude Score of 2.99 for Jose Cuervo.

*p<= 0.15

Patron has the average Attitude Score of 3.0, 1800 2.8 and Jose Cuervo 2.9.  Implications of this analysis are that Patron has the best consumer perception and brand liking, but is closely followed by Jose Cuervo and 1800. 1800 has the lowest consumer liking.

top

Please click here for Conclusions.

Back to Table of Contents

Conclusions

Correlated t-test

Jose Cuervo has the highest brand index score at 30.1 followed closely by Patron at 29.6 and 1800 at 27.9 respectively.  The former two are close to their brand index mean, which indicates parity among those two brands within the sample size, but respondents on average felt that 1800 ranked lower. The negligible difference between Patron and Jose Cuervo means that in the general population one could not accurately hypothesize which brand the general population prefers between them, but that one could consistently expect a preference for them over 1800. Implications of this analysis are that Jose Cuervo has the best consumer perception and brand liking overall.

top

Between-Groups t-test for Brand A, Patron

The brand index and ad index score for those respondents that were up-movers, pre to post exposure was higher for Patron relative to those who moved down.  This result is logical, and means that in 85 or more samples out of every 100 samples drawn from the same population as this sample, the mean brand index score for Patron, for the up and down movers would be about what it is in this sample. 

top

Chi-Square

Doing a chi-square allows a significance inference test for the brands.  24 out of 53 valid responses have a higher Brand A, Patron brand index score than Brand B, 1800 brand index score.  This is valuable information as we try to understand what results are indicative of the advertisement and what results are indicative of prejudgment. The correlation coefficient is a value between 1 and negative one.  The value from the data comparing Patron and 1800 brand index scores shows the correlation to be .8 which is a relatively strong correlation of statistical significance. From this result we can conclude that within the general public there is a slight tendency for people to prefer Patron to 1800. 

top

Frequency Test

After viewing the ad, approximately 50% felt differently about the brands, with the other half's opinion remaining unchanged. The brand that respondents felt the most differently about was Patron, at 26%, 1800 close behind at 20.8%, and Jose Cuervo had little change at only 5%. Of those who changed their opinion, 16 or 30.2% felt more favorable towards the brand, and 11 or 20.8% felt less favorable.

top

Frequency for Brand Index Score

The results obtained from this comparison of Brand Index Scores indicate Jose Cuervo has the highest brand index score at 30.1 followed closely by Patron at 29.6 and 1800 at 27.9. Jose Cuervo also had the highest frequency of top of the mind recall for the free-form answers in the beginning of the survey, with 21 respondents naming it. It can be concluded from these measurements that Jose Cuervo has the highest favorable brand perception among the market share that my survey was predominantly filled with: young twenty-something females.

top

Simple Correlation Coefficient

The correlation coefficient was found using only respondents who purchase alcohol three or more times a month.   Between Brand A, Patron and Brand B, 1800 only 33 of the original 53 respondents fit the criteria of being frequent purchasers, The correlation found is slightly higher here, at 0.9 instead of 0.8 as was the case using all the data, which is a logical conclusion, that the significance would increase among frequent users of the product.  

top

Regression Analysis

Regression uses a scatter plot on an x and y-axis, x being dependent variables and the y being the independent. An attempt is made to explain y by virtue of a line drawn between the points, and here the 10 Likert items of the survey (x’s) were plotted against each change score (y’s) in order to adequately explain the change score for each brand of tequila. The strongest connection for Patron was found on the final Likert question concerning the quality of the Patron brand. 

The positive slope suggests that respondents had an improved opinion after viewing the ad for Patron, and they also raised their opinion of the brand itself. 36% of people's Likerts scores can be used to predict a higher ad index score for this brand. However, in Table A, the F-ratio of 2.35 is not significant, and therefore the results cannot be projected onto the population.

The Likert item with the lowest significance in the coefficients table (less than 0.05) is the one most correlated to a higher ad index score.  In Patron’s case, this is the Likert question asking for a ranking of the belief that Patron uses higher quality agave plants to make their brand of tequila. 

For 1800, the positive slope also suggests people in the sample increased their opinion after viewing the ad for 1800, and their opinion of the brand itself. The percentage for 1800 is only 17%. The Likert question with the highest correlation ranked the belief that 1800 is stronger that other brands of tequila. For Jose Cuervo, the slope was also positive, with a percentage of 20%.

top

Discriminant Analysis

Discriminant analysis was run on the brand with the most movers, which in this case it was Patron tequila. These two groups of up and down movers were used as the categorical dependent variable and the brand attributes (10 Likert items) were used as the independent variables. This analysis was conducted to determine whether or not we can accurately predict, random chance accuracy aside, the divide between these groups through their rating of brand attributes or Likert responses. The largest value, taking the absolute value, was the first Likert item, which asks the respondent to score if they feel the ad for Patron is confusing. 

This discriminant analysis was only moderately successful in showing any difference between up and down movers for this brand.  While Wilks’ Lambda was significant, as indicated by the chi-squared, the group statistics table illustrates that there are no large differences between mean scores between groups.  Surprisingly, overall the Likert items which measured the reaction to the ad did not appear to have a strong correlation to the whether the respondent’s opinion of the brand overall would improve or decline over the course of the survey, with the exception of the first Likert question.

top

ANOVA

Analysis of variance between groups tests a null hypothesis that several group means are equal in the population, by comparing the sample variance estimated from the group means to that estimated within groups. Using the brand attribute “This is a good brand of tequila,” as a dependent variable, and the income and intent to buy this brand as two independent variables, the two-way factorial ANOVA test attempted to determine their relationship.

The F ratios were significant, therefore we can conclude that in 85 or more samples out of every 100 drawn from the same population as this sample of 53, we would expect to find similar means scores and variances of the same magnitude.  From this analysis we can conclude that people who think this is a good brand of tequila are more likely to make under $40,000.

MANOVA

This procedure tests any of a wide variety of null hypothesis about the effect of other variables on the mean value of several correlated dependen