Author: ir. Jos Deliën, April 5th – 2010
© Day Two & ir. Jos Deliën

Abstract
Recently, much attention in the SEO arena has been given to the full discovery of the proclaimed ‘200-plus Search Ranking Factors’ that Google uses, or might use. On the contrary, fairly little buzz has been spread around the subject of Search Ranking Factor Weights: the relative importance that Google assigns to each individual Search Ranking Factor. The lack of rumor among SEOs regarding analysis or even speculation about Search Ranking Factor Weights is stunning: if it would be known which ranking factors are to be considered “most important” then search marketing efforts can become more focused, less time-consuming and provide a more predictable ROI. All of these are objectives to be strived for, hence the need for better insights in the field of search engine optimization. However, the use of Search Ranking Factor Weights by Google appears to be dynamic instead of static, making efforts to determine actual values of each Search Ranking Factor Weight more problematic.
This article aims to attract more attention to Search Ranking Factor Weights in general, to introduce the concept of Conditional Factor Weights and a Conditional Factor Weights Function as a possible way to explain different types of SERP outcomes for specific search queries over time, to close the gap on missing Search Ranking Factors, and to explain the use of only one generic Search Ranking Model for all purposes instead of multiple ones for different situations.

Search Ranking Factor Weights: the basics
Regarding On-Page search engine optimization, much consensus exists on the relatively high importance of – for instance – the following factors:
a) Exact Match of Domain Name and Search Query;
b) Position of Keyword in Title Tag;
c) Occurrence of Keyword in H1 Heading Tag (preferably located near the page top).

For Off-Page optimization, the same general consensus applies to these (non-keyword related) factors:
d) PageRank of Page (representing the Global Link Popularity of that particular page);
e) Number of ILDs to Host Domain.

Assumptions, model 1:
- For simplicity matters, let’s assume a minimum score of 0 and a maximum score of 100 for each factor a - e;
- Factor Weights are not applied (or all equal to 1);
- No other Search Ranking Factors are applied;
- Let the end result of the Factor Score Outcomes be defined as: End result(i) = a(i) + b(i) + c(i) + d(i) + e(i);
- The web page with the highest end result score is granted the number 1 SERP position, the other pages are presented in descending order of the end result score;
- On the example search query “vanilla cake recipe” page X, Y and Z score the following factor outcomes, leading to a specific SERP ordering:

	Factor Score Outcomes
Page(i)	a	b	c	d	e	End result	SERP position
X	0	100	100	50	35	285	1
Y	0	50	100	75	50	275	2
Z	100	50	0	50	65	265	3

Assumptions, model 2:
- The same assumptions apply as in model 1 except for the introduction of Factor Weights;
- The Factor Weight of factor a is defined as α = 0,8;
- The Factor Weight of factor b is defined as β = 0,5;
- The Factor Weight of factor c is defined as γ = 0,2;
- The Factor Weight of factor d is defined as δ = 0,3;
- The Factor Weight of factor e is defined as ε = 0,4;
- Let the end result of the outcomes be defined as: End result(i) = αa(i) + βb(i) + γc(i) + δd(i) + εe(i).

	Factor Score Outcomes
Weights	α = 0,8	β = 0,5	γ = 0,2	δ = 0,3	ε = 0,4
Page(i)	a	b	c	d	e	End result	SERP position
X	0,8 * 0	0,5 * 100	0,2 * 100	0,3 * 50	0,4 * 35	99	2
Y	0,8 * 0	0,5 * 50	0,2 * 100	0,3 * 75	0,4 * 50	87,5	3
Z	0,8 * 100	0,5 * 50	0,2 * 0	0,3 * 50	0,4 * 65	146	1

As a result, the SERP ordering for the same particular query has changed dramatically from model 1 to model 2, making it somewhat more complex for SEOs to determine which factor score needs enhancement to reach a higher SERP ranking.
It’s also interesting to note that Google is able, without any adjustments to the Ranking Model, to “twist the knobs” of each Ranking Factor from time to time by decreasing the weight of one factor and increasing the weight of another, either manually or automated. But in order to process such adjustments in an automated manner, an additional – yet simple – function is needed, which will be introduced hereafter.

The introduction of Conditional Factor Weights
After studying the excellent SEOmoz article regarding an industry survey on possible Search Engine Ranking Factors and their relative importance, it may be noticed that a total absence of query-specific factors exists. However, well-known search parameters such as (for example) “Number of Page Results for Search Query” and “Temporal Search Volume for Search Query” – which apparently weren’t considered to be a ranking factor for the “normal” Search Ranking Model - might well be an integral part of the overall equation with the introduction of the concept of Conditional Factor Weights and an additional Conditional Factor Weights Function.

Let’s assume the existence of a Conditional Factor Weights Function that uses the values of one or more Conditional Factor Scores as an input trigger for altering the output values of Conditional Factor Weights. The direct result of that Conditional Factor Weights Function is that specific search queries over time make use of different Conditional Factor Weights than the same search query from another end-user’s location or timing of their search query and of course other search queries, and that the outcome scores of individual factors become more or less important for high ranking purposes.

This also means that only one overall and fully-automated generic Search Ranking Model and underlying algorithm can be held responsible for all types of SERP outcomes and SERP lay-outs (e.g. regular, One Box, Universal Search, et cetera) for the same search query, over time. The Query Deserves Freshness algorithm (QDF) for example, where established inbound links to specific web pages matching the search query are to be considered far less important for high rankings than the dependent factor Domain Crawl Rate, can be controlled by only two – partially independent – search query specific factor values: Temporal Search Volume for a specific Search Query, and Temporal New Citations for a specific Keyword. With the use of Conditional Factor Weights, the lay-out and order of SERPs for that specific search query can be temporarily changed in case one or both of those two QDF-factors reach a certain value, without having to use another Search Ranking Model than the generic model itself.

Let’s illustrate the theory described above with a simple additional case model:

Assumptions, model 3:
- The same assumptions apply as in model 2;
- The values ‘low’, ‘medium’, and ‘high’ are assigned as possible outcomes for the additional factor ‘f’: “Number of Page Results for Search Query”;
- The example search query “vanilla cake recipe” has scored the value ‘high’ for this factor.
- The following Conditional Factor Weight Function applies:

if ( f == ‘high )
{
}
else
{
}

In this particular case, the example factor score outcomes are as follows, again resulting in a change of the SERP ordering:

	Factor Score Outcomes
Weights	α = 0,2	β = 0,2	γ = 0,7	δ = 0,8	ε = 0,9
Page(i)	a	b	c	d	e	End result	SERP position
X	0,2 * 0	0,2 * 100	0,7 * 100	0,8 * 50	0,9 * 35	161,5	2
Y	0,2 * 0	0,2 * 50	0,7 * 100	0,8 * 75	0,9 * 50	185	1
Z	0,2 * 100	0,2 * 50	0,7 * 0	0,8 * 50	0,9 * 65	128,5	3

Conclusions

• With the introduction of Conditional Factor Weights and a Conditional Factor Weights Function a new light is shed onto search engine optimization in general and the hunt for a complete Search Ranking Factors overview in particular;
  

• If Conditional Factor Weights can alter the (relative) influence of other factors for specific queries, search engines in general and Google specifically might be able to cope with just one generic Search Ranking Model instead of multiple Ranking Models for different situations;
  

• Query-specific factors may be considered to be a part of Google’s ‘200-plus Search Ranking Factors’ model and the same may be true for local factors, temporal factors, citation factors and outcome values of non-search web usage statistics that were previously thought not to be a part of the “normal” Google Search Ranking Model;
  

• Outcomes of query-specific or other factors may result into different Conditional Factor Weights per search query, different total scores for individual web pages and different SERP orderings and lay-outs for that specific Search Query (such as One Box Results, Universal Search Results et cetera);
  

Suggestions for future research
It is suggested for future research – if reasonably possible at all - to statistically test the conceptual hypothesis presented in this article for general consensus rejection or acceptation. Machine-based simulation models may be used for that purpose in combination with temporal search volume shifts as presented in Google Trends. I further advise the use of SEOmoz’s LinkScape data for link factor analysis.