Why Is It Better To Be First? Search, Ranking, and Order Effects
We all know that, when it comes to search results, it’s an advantage to be positioned high on the list. Early items are given more attention and are chosen more often. Summary statistics indicate that the first item listed in a Google search gets a third of all clicks made (Dean 2019). Search order thus has an enormous potential to influence decisions — for example, when internet search rankings were manipulated for political candidates, the candidate given a more prominent position in search results was also voted for more often (Epstein & Robertson 2015).
But why this is the case — why exactly it is an advantage to get a higher search ranking and end up earlier on a list of results?
There are a number of explanations for the observed value of a high ranking, ranging from a perception of search rank as an indication of quality to the comparisons a particular search order invites us to make. Here, I consider some possible explanations and discuss their relative merits in relation to search engine results.
The dominant explanation — at least for a majority of internet searches — is that search ranking is seen as a good indication of the quality of the search hit. Items higher up in a list are a better match to the search, and include the information required by the searcher. Search ranking acts as a cue, strongly steering participants’ selections (e.g. Pan et al 2007). For most searches this is a perfectly adequate strategy: if the first hit and the first click gets you what you want, why would you continue looking?
There is, however, a subset of searches where this is not enough. The study cited above on choosing political candidates is a case in point, as are information searches in areas where the quality and type of information are paramount: Should I vaccinate? How do you make the best garam masala? How secure are different horse-riding safety vests?
In addition to web users’ perception of search ranking as an indication of quality, the fact that search pages most often are organized as lists must matter. We tend to read lists from top to bottom, so even if search rank is not used as a quality cue it is to be expected that earlier items are first attended to. If we allow attention (as well as time) to be a limited resource, it follows that a long list, such as the ones produced by search engines, will have to be truncated at some point. There will only be enough time and capacity for a certain proportion of that list. Naturally it is an advantage to be among the options given attention, and the earlier an option appears on a list, the more likely is it to be included among the items attended to. This does not seem to be enough to understand the prominence of first position, however. Even when several options are attended to, earlier options have an advantage, as in the case of the (manipulated) search for political candidates described above.
Such a finding could be explained if we allow the person reading the list to gradually discover how much time and attention they want to spend on the search results. A tendency to overestimate one’s time and cognitive capacity implies that later items receive less attention as these resources run out. With this we only need the extra assumption that more attention is advantageous to explain why search rank matters so much. Attention will be an advantage in so far as receiving attention leads an option to be more preferred, perhaps because more reasons for favouring that item have been generated (for a strict test of this assumption cf. Pärnamets et al 2015).
That being first (or last) on a list is an advantage was established by Ebbinghaus (1885) through his discovery of primacy and recency effects. Serially presented items that occur in the first or last positions are best remembered, resulting in a u-shaped curve — the serial position curve. Several mechanisms in turn lie behind this effect: early items have been rehearsed more often by someone trying to remember the list, items in the middle often get less attention (and are therefore less likely to be stored in memory) and the last items have the chance of staying in short term memory until the participant is asked to recall the list. If more items are added, this will push out earlier content in the limited short-term memory, so that the last presented items are most likely found there. Note, however, that in order to test recall, the items in front of the participants (the list) are removed. If items remain in front of the participant they do not have to be recalled. Memory decay could thus explain why participants would go back to and prefer early items after having scrolled down so that first hits are no longer visually available, but it is an unlikely explanation for differences between the highest search rankings. It is also important to note that the terminology of primacy and recency effects can be applied to any finding in which early or late items are given an advantage, but unless the effects are linked to explanations (e.g. rehearsal attenuates memory decay), the terminology is nothing more than a re-description of an observed pattern.
It seems as if something more is needed to understand why items with high search rankings are preferred. A good candidate explanation is that the order in which options are presented on a list influences how they are compared to one another. When making relative judgments, we tend to emphasize the unique features of the option we are focusing on (Tversky 1977). This means that as you move from an earlier to a later option on a list, you will focus on the special features of that later option when evaluating it — those that set it apart from the previous option. If the features are unique and positive, the latter option is perceived as positively differing from the previous one. If they are unique and negative, the option instead differs in a negative way. For the options on a list, some features will overlap and others will not. The focus on uniqueness entails that features that are shared between two subsequent options will be given less weight. If options are evaluated on positive features (reasons for choosing a particular alternative) this should mean that for each additional option identified, more and more weight will be placed on the remaining distinguishing features of those alternatives. This gives later options an advantage since unique and uncommon features will get more attention, escalating positive judgements. Similar phenomena have been observed for a range of competitions, from synchronized swimming (Wilson 1977) to Eurovision song contests (Bruine de Bruin 2005). If options instead are evaluated on negative features (reasons for not choosing a particular alternative) judgements will instead de-escalate and become increasingly negative (Bruine de Bruin & Keren 2003).
In all likelihood all of these mechanisms matter for how search engines are used. The mechanisms become even more important as advertisements and other content are added to search pages. Note, for instance, that few participants are able to reliably identifying advertisements on search pages (Lewandowski 2017). This means that presenting a political ad over a political search may bias not only the clicks different hits receive, but also how latter options are evaluated in relation to earlier ones.
Search order matters: not only because search rank is equated with qualitative content, but also because it influences the amount of attention an option gets, and how a series of options are compared to one another. Manipulating search order has an enormous potential to influence our judgments and our decisions when we’re online as well as off.
