Pascale G. Quester PhD (University of Adelaide)

Cam Rungie (University of South Australia)

 Abstract

Studies in sponsorship have only recently aimed to assess empirically the effectiveness of this communication method on target audience awareness. A telephone survey of relatively large samples before and after the 1994 Adelaide Formula One Grand Prix, involving 297 and 206 respondents respectively, generated data which was analysed using Hierarchical Log Linear Analysis in order to assess the suitability of this approach in similar instances of marketing research. The procedure followed is described in details and compared with a more traditional method of analysis and was found to procure an elegant and effective alternative.

INTRODUCTION

The increasing proportion of promotional budgets allocated by companies to sponsorship programs has resulted in a recent surge of academic and practitioner’s research in this area (Meenaghan, 1994). Sponsorship activities are often undertaken with a variety of objectives in mind (Abratt et al, 1987, Hoek et al, 1993) but one often quoted corporate aim by sponsors is the achievement of target market recognition and awareness (Crowley, 1991, Cegarra, 1994).

Attempts at assessing the impact of a number of sponsorships on target audiences’ perceptions have faced limitations as a result of the many contributing variables to levels of awareness (Couty, 1994) with the most reliable studies using before and after measurements and relatively large samples in order to ascertain whether the event has resulted in any significant changes.

This study, the results of which are used for this analysis, involved two consecutive waves of telephone survey conducted immediately before and immediately after the 1994 Adelaide Formula One Grand Prix. It aimed to quantify the effect of the passing a sporting event over the public perceptions of its sponsors and to explore the incidence of ambush-type effect over these perceptions. Whilst the general findings of the study are reported elsewhere (Quester 1997a; Quester 1997b), the application of a specific statistical procedure to the data from one such experiment constitutes the focus of this paper.

Approximately two hundred and fifty respondents took part in each of these surveys and were asked to indicate whether they recognised any of the 22 company names listed to them as acting sponsors of the event. Amongst the 22 names, 13 were official sponsors, 5 were supplier-sponsors and 5 represented "control names" used for the purpose of measuring the incidence of the sort of ambush marketing reported in several other studies (Sandler and Shani, 1989, Meenagahn, 1994).

The paper is organised as follows. The next section presents the main findings of some of the literature dealing with sponsorship. The results relevant to one of the companies listed in the study are then presented. A subsequent section of this paper describes how hierarchical log linear analysis was applied to the data in order to determine the model that could best predict the observed awareness levels. The loglinear approach was then compared with the more traditional method of analysis for all companies involved in the study. The paper concludes with a number of brief observations and points to future research directions in this area.

SPONSORSHIP RESEARCH FINDINGS

In addition to studies dedicated to the descriptive assessment of the magnitude of Sponsorship budgets on a national or global basis ( Meenaghan 1991; CEASA Report 1991), a number of authors have attempted to categorise sponsorship activities, mainly on the basis of the characteristics of the sponsee i.e Sports, Art, Events, Causes and, more recently, television broadcast programs. Sponsorship has subsequently been the object of focus of more specific studies, dealing with one form or another.

One interesting direction of previous research has involved the analysis and discussion of the range of organisational objectives best served by the use of Sponsorship. A survey of 140 large organisations in the UK, undertaken by Witcher et al. (1991) provided an indication of the most preferred form of sponsorship and the rationale behind this choice. More than half of the respondents had supported professional sports. The main objective stated sponsorship activities was, for a large majority of two thirds, the promotion of company image for both Sports and Art sponsorship. Functionally, however, Sports tended to be the domain of the Marketing department or managers while Art was used mainly for Public Relation purposes and was therefore placed in the area of responsibility of the Public relation manager or the relevant consultant agency.

With particular reference to sports sponsorship, an earlier study by Abratt et al. (1987) demonstrated that the corporate objectives were articulated around three main dimensions, the potential media (television) coverage, the promotion of corporate image and the potential to attract spectators and audience as customers. These rather more specific and measurable objectives were obtained by a self administered questionnaire mailed to some 60 South African sponsor firms.

Cause-related Sponsorship, on the other hand, was argued by Varadarajan et al. (1988) to be motivated by a range of objectives from gaining national visibility to pacifying customer group or counter negative publicity and could include such profit-based objectives as increasing sales or market share, repeat or trial purchases. Based on a number of case study examples, the authors categorised such sponsorship activities as part of the broader cause-related marketing for which they developed a framework on the basis of the number of participating brands and firms, the level of association, the geographic scope and the strategic or tactical nature of the objectives.

More recent studies have confirmed the variety of objectives sought by sponsors, making sponsorship a most versatile and flexible marketing communication tool (Farrelly et al. 1997; Quester and Farrelly 1997).

In addition to understanding the sponsors' objectives, some authors have attempted to explain the phenomenal growth of Sponsorship itself. Suggested reasons included the following (Meenaghan, 1983): escalating cost of advertising media, increasing inefficiencies in existing media (clutter, zapping), increasing media coverage of sponsored events, emerging opportunities resulting from an increase in leisure activities and government policies with regards to advertising of tobacco and alcohol products.

Other potential justifications could of course be hypothesised, such as the introduction of mechanisms for the negotiation of such agreements, the active prospecting by potential sponsees of the potential sponsors, the diffusion of sponsorship as a marketing tools through management although not yet presented by any authors dealing with the subject. Cornwell (1995) even suggested an entirely new perspective on the issue, whereby marketing strategy is designed around sponsorship programs, building differentiation and competitive advantage through ‘sponsorship-linked marketing’.

A somewhat different perspective on Sponsorship is provided by these authors who have focused on consumers' perceptions and acceptance of sponsorship activities. One such study by Marshall (1992) reported on the different attitudes towards sponsorship amongst British, German, Spanish, French and Italians respondents. This study found that sponsorship was perceived as similar to advertising, and indicated that respondents could identify the commercial purpose of the communications to which they were exposed. In addition, the findings supported the appropriateness of sponsorship for the purpose of building corporate image as most respondents agreed that "sponsorship promotes a good image for the sponsor and its products " and a large majority stated that the image of the sponsor was either enhanced or unchanged as a result of the sponsorship activity. Moreover, sponsor benefited from consumers' perceptions that they provided vital funds for organising events.

Marshall also investigated the issue of awareness and retention by studying the awareness levels over time for a number of brands and events. He found that decay for the Mars brand after its involvement in the London Marathon race between 1986 to 1988 did not exceed 14 %. Awareness, however, was also found to be influenced by the extent to which the organisation "multiplied" the sponsorship effect by supporting its effort with additional communication activities, such as advertising.

Similarly, Hitchen (1994) reported on data compiled by ISL Olympic, including the result of a two-wave, four-country study conducted in 1988 (USA, France, UK and Brazil) and a three-wave, three-country study conducted in 1992 (Spain, UK and the USA). Overall, 89% of respondents (they were approximately 500 per wave) identified correctly the commercial objectives of the sponsors with 82 % agreeing that the sponsor's financial support was critical to the event.

Not all of these results are undisputed, however. For example, some authors, including Hastings (1984), have denied that sponsorship effects are the same as those of advertising. Additionally, the event of the ‘over-commercialised’ Atlanta Olympic Games have contributed to a somewhat more cynical public perception of sponsors’ objectives and of sponsorship itself.

One interesting development in the area of sponsorship, Ambush Marketing, refers to the situation where competitors of official sponsors design their communication strategy in such a way as to create the impression that they too are involved in sponsoring the activity, sport or event, concerned. The term was coined by Bayless (1988) with respect to special-event sponsorship as this tactic was used by some US companies during the 1988 Olympics. Indeed a distinct definition was developed to differentiate it from official sponsorship (Sandler and Sani, 1989):

Sandler and Sani evaluated the effectiveness of Sponsorship and Ambush Marketing for the 1988 Winter Olympic Games and found that official sponsors were correctly identified in only four out of seven product categories. In all cases, being an ambusher provided better recall than for those companies which did not engage in any event related effort. In other words, companies intending to chose this cheaper alternative to that official sponsorship seem to have nothing to lose.

As noted before, there is still relatively little research available in the area of sponsorship and few studies that investigate the particular area of Ambush Marketing, despite increasingly visible occurrences of it: Nike has been particularly bold in its ambush tactics. While not involved in any official sponsor capacity with these events, it placed its highly recognisable logo on a building --purchased for this very purpose-- across from the main location of the Atlanta Games. Similarly, by sponsorship 6 teams of the Soccer World Cup team (including Brazil and the Paris-St-Germain clubs), Nike is also gaining maximum exposure during this world wide event, building a special site in Paris which it operates jointly with the PSG and where merchandise bearing its logo are sold. While some research may be undertaken by ambushers and ambushed marketers, they are kept under the seal of secrecy, preventing a critical appraisal of their methodology. As a result, few academic researchers have taken the challenge of exploring the issue in Australia, a country where sports and sponsored activities are particularly popular.

The next section presents the results relevant to one of the companies listed while the following section of this paper describes how hierarchical log linear analysis was applied to the data in order to determine the model that could best predict the observed awareness levels. A comparison follows with the more traditional method of analysis for all companies involved in the study. Finally the conclusion provides a number of brief observations and future research directions.

THE CASE OF ELECTRONIC DATA SYSTEMS

Only one of the hierarchical log linear analyses conducted for the study is presented in this section, based on data relating to Electronic Data Systems, one of the major sponsor of the 1994 Australian Formula One Grand Prix for which data was collected both before and after the event took place. The analysis fitted a model to the three variables presented in Table 1.

As Table 1 shows, the variables were:

Aided awareness of Electronic Data Systems (Variable Q2.15)

1 = Yes, aware

0 = No, not aware or don’t know.

Survey sample (Variable TIME)

1 = Pre Grand Prix survey sample

2 = Post Grand Prix survey sample

Attended any Adelaide Grand Prix (Variable AT_ANY)

1 = Yes, have attended an Adelaide Grand Prix

2 = No, have not

The basic results, therefore, without totals, constitute a 2 X 2 X 2 table with eight elements, on which hierarchical loglinear analysis was undertaken in a manner detailed in the following section.

Application of hierarchical loglinear analysis to EDS

Let the original table be represented as a matrix A with elements a_ijk where:

i = 1,0 representing aided awareness of EDS (1 = yes, aware. 0 = no.)

j = 1,2 representing time (1 = pre and 2 = post survey sample.)

k = 1,2 representing attend any Adelaide Grand Prix (1 = yes, attend. 0 = no, did not attend.)

The original saturated model was:

This model had 27 parameters. Since Table 1 had eight elements, 19 of the parameters were redundant and were automatically set to zero. These parameters thus treated were said to be ‘aliased’.

Hierarchical log linear analysis was then used to examine the impact of setting the remaining parameters to zero with the higher order interactions examined first. The final model from the analysis, as shown below, was:

 Thus the figures in Table 1 were predicted by using parameters for

(1) overall awareness for EDS

(2) the change in sample size between those who attended and Adelaide Grand Prix and those who did not

(3) interaction between (1) and (2)

That is, the pre and post samples categories were not specifically considered in predicting the results shown in Table 1. The analysis presented here is an edited and annotated version of the printout from SPSS for Power Macintosh.

* * * * * * * * H I E R A R C H I C A L L O G L I N E A R * * * * * * * *

DATA Information

503 unweighted cases accepted.

0 cases rejected because of out-of-range factor values.

0 cases rejected because of missing data.

503 weighted cases will be used in the analysis.

FACTOR Information

Factor Level Label

Q2.15 2 Electronic Data Systems Australi

TIME 2 Survey Sample

AT_ANY 2 Attended Any Adelaide Grand Prix

The analysis commenced by confirming that there were 503 cases and three variables, each of which had two possible values. The table being modelled had 8 elements (2X2X2).

The analysis commenced with a fully saturated model. Each of the elements in the table was predicted using all three of the factors and all interactions between the factors.

The Iterative Proportional Fit algorithm converged at iteration 1.

The maximum difference between observed and fitted marginal totals is .000

and the convergence criterion is .250

Goodness-of-fit test statistics

Likelihood ratio chi square = .00000 DF = 0 P = 1.000

Pearson chi square = .00000 DF = 0 P = 1.000

 As expected the model fitted perfectly (a model is considered fitted if the test statistic, Chi square, is not significant). The expected values for the table generated by the model did not differ significantly from the observed, actual values, as shown by the SSPS print out reproduced overleaf.

 Tests that K-way and higher order effects are zero.

K DF L.R. Chisq Prob Pearson Chisq Prob Iteration

3 1 .260 .6103 .260 .6104 2

2 4 6.634 .1565 6.542 .1621 2

1 7 25.357 .0007 26.256 .0005 0

Tests that K-way effects are zero.

K DF L.R. Chisq Prob Pearson Chisq Prob Iteration

1 3 18.723 .0003 19.714 .0002 0

2 3 6.375 .0947 6.283 .0986 0

3 1 .260 .6103 .260 .6104 0

 A preliminary diagnosis was undertaken which examined if the expected values would still fit the observed values in the table if the factors in the model did not fully interact. The non- significant results indicated that it would be possible to fit a simpler model which did not include all the interactions.

Backward Elimination (p = .050) for DESIGN 1 with generating class

Q2.15*TIME*AT_ANY

Likelihood ratio chi square = .00000 DF = 0 P = 1.000

At this point the hierarchical model building commenced, with a restatement that the starting point was a fully saturated model which fitted perfectly.

If Deleted Simple Effect is DF L.R. Chisq Change Prob Iter

Q2.15*TIME*AT_ANY 1 .260 .6103 2

Step 1

The best model has generating class

Q2.15*TIME

Q2.15*AT_ANY

TIME*AT_ANY

Likelihood ratio chi square = .25967 DF = 1 P = .610

 The hierarchical process identified that the full three-way interactions could be removed without creating a significant difference between expected and observed values in the table. There were three two-way interactions, all of which, at this point, were in the model.

If Deleted Simple Effect is DF L.R. Chisq Change Prob Iter

Q2.15*TIME 1 .946 .3307 2

Q2.15*AT_ANY 1 5.432 .0198 2

TIME*AT_ANY 1 .106 .7444 2

Step 2

The best model has generating class

Q2.15*TIME

Q2.15*AT_ANY

Likelihood ratio chi square = .36594 DF = 2 P = .833

 The process examined the possibility of eliminating some of the two-way interactions. The following print out extract shows that one was dropped.

If Deleted Simple Effect is DF L.R. Chisq Change Prob Iter

Q2.15*TIME 1 .891 .3452 2

Q2.15*AT_ANY 1 5.377 .0204 2

Step 3

The best model has generating class

Q2.15*AT_ANY

TIME

Likelihood ratio chi square = 1.25709 DF = 3 P = .739

 Subsequently, another two-way interaction was dropped as shown below.

If Deleted Simple Effect is DF L.R. Chisq Change Prob Iter

Q2.15*AT_ANY 1 5.377 .0204 2

TIME 1 .002 .9646 2

Step 4

The best model has generating class

Q2.15*AT_ANY

Likelihood ratio chi square = 1.25905 DF = 4 P = .868

 Then, somewhat more surprisingly, time was dropped as a factor, as there appeared to be no need in the model to differentiate between the pre and post Grand Prix survey samples.

If Deleted Simple Effect is DF L.R. Chisq Change Prob Iter

Q2.15*AT_ANY 1 5.377 .0204 2

Step 5

The best model has generating class

Q2.15*AT_ANY

Likelihood ratio chi square = 1.25905 DF = 4 P = .868

 Following this last step, however, the process concluded that no more components of the models could be dropped without significantly reducing the fit.

The final model has generating class

Q2.15*AT_ANY

The Iterative Proportional Fit algorithm converged at iteration 0.

The maximum difference between observed and fitted marginal totals is .000

and the convergence criterion is .250

Goodness-of-fit test statistics

Likelihood ratio chi square = 1.25905 DF = 4 P = .868

Pearson chi square = 1.25776 DF = 4 P = .869

 The final model can therefore be restated as including the two factors (1) awareness of EDS and (2) attendance to any Adelaide Grand Prix as well as the interaction between the two factors.

Hence, the final model was:

 Where the parameter estimates were:

 Aliased values were automatically set to zero due to redundancies in the model. Thus the final model required the estimation of four parameters only. The calculation of the expected values for each of the eight elements in the original data table are shown in Table 2. The expected values were calculated by totalling the appropriate constants and then taking antilogs. The loglinear model showed that the differences between the observed and expected values in Table 1 were not statistically significant, using a Chi-Square test.

Table 2 Calculation of expected values from the Model

Table

Model

Elements

Observed

Parameter values

Total

Expected

Aware

Attend

Time

values

Constant

Aware

Attend

Time

Aware*Attend

values(a)

55

3.882

0

.117

0

0

3.998

54

54

3.882

0

.117

0

0

3.998

54

53

3.882

0

0

0

0

3.882

49

44

3.882

0

0

0

0

3.882

49

61

3.882

.567

.117

0

-.42

4.143

63

65

3.882

.567

.117

0

-.42

4.143

63

82

3.882

.567

0

0

0

4.449

86

89

3.882

.567

0

0

0

4.449

86

503

Total

503

  Furthermore, a normal probability plot for the differences between the expected (ie. from the model) and observed values in the table showed a reasonably straight line, indicating that the normality assumption for the Chi-Square test was valid. Additional assumptions of the Chi-Square test were that 1) no cell in the table exhibit an expected value of less than 1 and 2) that very few cells show expected values of less than 5. These assumptions were upheld as can be seen in Table 2.

ANALYSIS FOR ALL COMPANIES IN THE STUDY

The two following analyses were then undertaken on each company separately. First, each of the two factors, (1) pre and post survey sample and (2) attended any Adelaide Grand Prix, were each cross-tabulated with ‘aided awareness’ for the company, producing two separate tables similar to Table 1 above. A Chi-Square test for independence was undertaken on each of the two tables, generating the P-values presented in Table 3. In addition cross-tabulations between ‘attended this Grand Prix’ and ‘aided awareness’ were generated and a further Chi-Square test undertaken. The P-values for these tests are also presented in Table 3. The sample size for the test on those attending this Grand Prix was 251. Table 3 also shows those values which were significant, that is those values less than or equal to .05.

Loglinear analyses were then undertaken for each company the results of which are summarised in Table 4. Clearly, they are extremely similar to the results in Table 3 with Castrol the only company for which the loglinear analysis identified additional significant factors. Otherwise the results were identical.

Table 4 - Significance : Loglinear Models

The loglinear models therefore were predicting a three way table. It is standard for loglinear analysis to consider all relevant lower order interactions and for factors to be significant for those cases where a higher order interaction is significant.

CONCLUSIONS AND DIRECTIONS FOR FUTURE RESEARCH

Several observations can be made on the basis of these results:

(1) Aided awareness was always significant. This indicates that the number of respondents aware often did not equal the number of respondents not aware; not a surprising result but one that is encouraging to sponsors aiming to achieve an awareness objective via sponsorship.

(2) Time was not always significant. This is also not surprising since, by design, the number of responses in the pre sample was close to the number in the post sample. Time was only significant when there was a significant interaction with aided awareness.

(3) Attendance was only significant when there was a significant interaction with aided awareness

(4) Time and attendance did not interact (except for Castrol)

(5) There were no three-way interactions. Aided awareness, time and attendance did not simultaneously interact. The effects of time and attendance were, therefore, independent.

Compiling Tables 3 and 4 involved twenty-two separate analyses. It is to be expected then, that there would be a small number of significant results produced by chance alone. Considering that 66 tests were undertaken to generate Table 3, three significant results could therefore have been expected by chance. However, there were twenty significant results, providing evidence to support the significance results shown in Table 2. The one exceptional result in the loglinear analysis (for Castrol) should not be overly emphasised, due to the number of tests involved. Castrol presents a number of specific characteristics which set it apart. In addition to its long standing involvement in sports car racing, its logo has become part of the paraphernalia of any ‘petrol head’. As a result, while the company was not a sponsor, merchandise displaying its logo would have been found for sale everywhere around the site as if it had been involved in an official sponsoring capacity. In addition, T-shirts and caps bearing the logo would have been very common in the audience as well, multiplying the exposure gained by the logo. In other words, Castrol prominence in the car racing world might have made it a ‘quasi-sponsor’ in the eyes of the public.

Researchers in marketing and other fields may benefit from further applications of Hierarchical Log Linear Analysis when trying to ascertain the extent to which interactions exist between variables to be included in a model. The example in this paper demonstrates the use with which such analysis can be conducted and provide clear results in the case of one research area of interest to academics and practitioners alike. It is hoped that future developments in this field may enable the modelling of other marketing phenomena when sample data can be collected that enable such statistical analysis.

Abratt R., Clayton B. and Pitt L." Corporate Objectives in Sports Sponsorship", International Journal of Advertising, 6(4), 1987, 299-311.

Bayless A. ,"Ambush Marketing Is Becoming Popular Event at Olympic Games", The Wall Street Journal, Feb 8, 1988.

Cegarra J., " La Place du Sponsoring dans la Strategy marketing de l’entreprise", Proceedings of the AFM/CERIAM Research Seminar, March 1994, Chambery (France), 1-11.

CEASA, Media Expenditure Report, 1991.

Cornwell B., "Sponsorship Linked Marketing Development", Sports Marketing Quaterly, IV, 4, (1995), 13-24.

Couty F., "L’evaluation de la notoriete du sponsoring sportif", Proceedings of the AFM/CERIAM Research Seminar, March 1994, Chambery (France), 33-46.

Crowley M. " Prioritising the Sponsorship Audience", European Journal of Marketing, Vol 25, no 11, 1991, 11-21.

Farrelly F., Quester P.G. and Burton R., "Integrating sports sponsorship into the corporate marketing function: An international comparative study", International Marketing Review, 14(3), 1997, 170-182.

Hastings G. "Sponsorship Works Differently from Advertising", International Journal of Advertising, 3(2), 1984, 171-176.

Hitchen A., "Sponsorship Gold at the ‘92 Olympics", ESOMAR Proceeding of the seminar How Advertising and Sponsorship Work, 1994, 119-138.

Hoek J., Gendall P. and Sanders J. " Sponsorship management and evaluation : Are Managers’s assumptions justified?", Journal of Promotion Management, 1(4), 1993, 53-66.

Marshall D. and Cook G, "The Corporate (Sports) Sponsar", International Journal of Advertising, 11, (1992) , 307-324.

Meenaghan T., "Commercial Sponsorship", European Journal of Marketing, 17(7), 1983, 1-75.

Meenaghan T. , "The Role of Sponsorship in the marketing Communication Mix" International Journal of Advertising, 10, 1991, 35-47.

Meenaghan T., "Point-of-View: Ambush Marketing :Immoral or Imaginative Practice?", Journal of Advertising Research, 34 (5), 1994, 77-88.

Quester P.G. (a), "Awareness as measure of sponsorship effectiveness: The Adelaide Formula One Grand Prix and evidence of incidental ambush effects", Journal of Marketing Communication, 3(1), 1997, 1-20.

Quester P.G (b), "Sponsorship Returns: Unexpected findings and the value of naming rights", Corporate Communications: An International Journal, 2(13), 1997, 101-108.

Quester P.G. and Farrelly F., ""The impact of Atlanta Olympic games sponsor: A Comparative study of South Australian and Victorian Audiences", Proceedings of the ANZMEC Conference, Melbourne, December 1997.

Sandler D. and Shani D. "Olympic Sponsorship vs Ambush marketing", Journal of Advertising Research, September 1989, 9-14.

Varadarajan P. and Menon A, "Cause Related Marketing: A Coalignment of Marketing Strategy and Corporate Philanthropy", Journal of Marketing Research, 53, 1988, 58-74.

Witcher, B., Craigen, G., Culligan D. and Harvey A., "The Links between Objectives and Function in Organisational Sponsorship", International Journal of Advertising, 10, 1991, 13-33.