Sample Empirical Results

• Link Design and Copy Testing. An unpublished experiment illustrates this technique. The authors compared two banner ads for a national dating service, but also tracked what the visitors who clicked on those banners did on the advertiser's site. The banners were identical in all design elements, including the name of the firm and a simple logo, except for the copy. Message A simply said 'Click Here' while the longer message B said 'Done with bars and time wasting classifieds? Serious about finding your soulmate?'

Table 11-1 shows that while message A won the click-through battle ( c ² = 147.353, p = .001), it lost the battle for subsequent page requests on the dating service's Web site (Poisson Regression, t = 8.321, p < .001). While it is tempting to pick the banner with the highest click-through probability, that banner performed worse in an absolute sense; it yielded fewer page views on the advertiser's site.

Table 11-1: Results of ad banner click-throughs and follow up page requests

	N	Banner Click-through	Follow Up Page Requests on the Site
Message A	2498	.1137	.0246
Message B	2361	.0779	.2134

The Elaboration Likelihood Model may explain these results (Petty & Cacioppo, 1986). The short copy and more peripheral route processing achieved higher click-throughs, but the long copy and central route processing lead to significantly more clicks on the advertiser's site.

• Opt-In E-mail Design. E-mail newsletters and other outgoing e-mail messages resemble mass media. They are marketer, rather than customer, initiated. Yet even though e-mail is the most popular Internet application in business and at home (Ramsey, 2001), for young (Pastore, 2002) and for old (BBCNews, 2002), there seems to be little published research on this useful customer relationship management tool (Krishnamurthy, 2001).

An early experiment tested personalized vs. non-personalized greetings in a hotel e-mail campaign. Against expectations, there were significantly better results with the nonpersonalized greeting (Marinova et al., 2002). The authors suggested that as the hotel obtained most names at registration, there was a weak relationship and recipients may have objected to the personalization.

In 2002 unpublished study, Hofacker and Voorhees of the Florida State University performed a simple experiment that compared three versions of an e-mail newsletter for a leisure and recreation company. One version was plain text, one used basic HTML, and another HTML version included graphical banner ads rather than simple text links. Contrary to expectations, the text condition out-performed both HTML versions, although the effect was slight .

• Page Design. The authors executed several studies of click-throughs on links in simple Web menus . One of the first studies utilized a marketing practitioner site that had, on its home page, eight links arranged in a four-row by two-column menu. The first hypothesis was a monotonic relationship between row and click-through probability. In other words, the higher on the screen, the higher the click-through rate would be. Secondly, given English reading habits, left to right and top to bottom, links in the left column should garner higher click-through rates than links in the right column.

Should a Web master put the two most important links as the first two items either in the left column or as the two items in the top row?

The experiment used a Latin Square design that rotated each of the links through each of the menu positions. Figure 11-5 shows the empirical results. Click-throughs for the individual positions show that the upper left location was the most clicked position, as predicted , but that the lower right location was second. English language reading habits suggest reading the upper left first, and the lower right, last. These results, however, reflect a primacy and recency effect. We replicated these results on other sites, with different length menus and different menu structures. The results suggest that given a goal of increased clicks on particular links, a Web master should place the second most important link last, not second.

• Site Design. Hofacker and Murphy (2000) also studied pages containing one or two links in order to test two propositions concerning online choice. They used two pages, i and j, with page i containing a superset of the links on page j. In other words, page i had all the links that page j had, plus an additional link or links. The first proposition was that the total clicks for page i would not be less than the total clicks for page j. The second proposition was that individual clicks on any link included in both pages i and j would not be less for page j than for page i. The data confirmed these simple notions.

  
  Figure 11-5: Click-through probability as a function of item position within a menu