Introduction

In 2003 I took my first steps in memetics. I had been trained as a master in linguistics and needed a subject for my master's thesis. I had stumbled on memetics a couple of years earlier and had already written a few papers on it for various assignments, hereby further exploring this young field and laying out a path for myself that was not so obvious for a trained linguist. At first sight linguistics and memetics have few things in common so the combination of these two was new.

I had the advantage of having one of the early meme-specialists, Dr. Francis Heylighen, right here at our university. Although he had not been working in the field anymore for quite a few years, his knowledge was still very accurate and he agreed to be my counselor for the writing of this paper. Together we agreed that it would be a fine idea to tackle some of the critiques and challenges that lay upon the field of memetics, mainly the shift from a purely theoretical to a science with practical implications.

We decided it was necessary to quantify the science of memetics by creating a small study that would show the basics of this methodology. We used a survey to question the survey-takers about certain memetic selection criteria. We then compared the results of the various criteria to the degree of spreading of the memes that were used in the survey.

In this study we surveyed 200 people and asked them to score various memes on the different selection criteria that were used (Novelty, Simplicity, Danger, Benefit, Authority and Replication Pressure). To make the various memes comparable we limited ourselves to the use of virus hoaxes.

The full thesis is presented here as a downloadable pdf file for those of you who want to read the full thing. Please use this reference to quote the thesis:

Chielens, K. (2003). The Viral Aspects of Language: A Quantitative Research of Memetic Selection Criteria. Unpublished master's thesis, Vrije Universiteit Brussel, Brussels, Belgium.

Hoaxes Used in the Research

The survey focuses on six different hoaxes. These hoaxes were selected using the lists of virus hoaxes as provided by various security software companies. For this research the websites of Sophos, Symantec, Mc-Afee and F-Secure were used.

Initially, eleven hoaxes were chosen based on information about their current spreading (as available on the aforementioned websites). These hoaxes were subjected to research on three major search engines: AltaVista, Alltheweb and Google (both web & Usenet search).

From the combined results of all search engines the average of the natural logarithm was taken and it was this final result that was used to make up the ranking of the hoaxes. The results of the eleven chosen hoaxes were then compared to find six hoaxes which had enough variation to test several grades of success in spreading. A correlation needed to be found so it was decided that it would be best to make three groups of hoaxes, a high spreading one, a middle spreading and a group of hoaxes that was not very successful but still successful enough to be used in quantitative research without having to classify the results as 'not trustworthy due to a too small sample'.

To avoid interference of two or more hoaxes over each other in the survey, the hoaxes were also filtered based on their content and methodology. For example the virus hoaxes sulfnbk.exe and jdbgmgr.exe both scored within the top three results of virus hoaxes found online but they contain the same type of warning; the first one targeting sulfnbk.exe, the second one demanding to delete jdbgmgr.exe. It is likely that if these two hoaxes would both have been included in the same survey, they would have a strong influence on each other as far as criteria such as novelty are concerned.

The hoaxes that were selected to be tested in the end were: jdbgmgr.exe, Good Times, Penpal Greetings, WTC Survivor, California IBM and Buddylst.

The Criteria Used in the Research

The criteria that were chosen to be included in the survey needed to be easily comprehendible by the average survey-taker. From Heylighen's list, six criteria were chosen. These criteria are novelty, simplicity, danger, benefit, authority and replication pressure. In the introduction to the survey each of these criteria were described to give the survey-taker an insight on the meaning of each criterion.

The descriptions of the criteria were repeated with every question in the survey, with the extra information of the values that could be entered for the criterion (a scale from one to five).

For example: "How easy is it to understand this message? Is it hard to grasp or is it pretty clear and simple? (1: Very Hard / 5: Very Easy)"

The criterion 'novelty' was renamed in the online survey as 'originality' this was done in order to avoid a misunderstanding or confusion with the concept 'new'.

Simplicity was not changed. Although dependant of the previous knowledge of the survey-taker this is a criterion that can easily be scored. The different background of the participants is compensated by the fact that each survey taker will score all hoaxes with the same previous knowledge.

The two criteria danger and benefit are both sub-criteria from what Heylighen called utility. By separating these two it can be verified whether or not one of these two is more important.

The authority selection criterion should be one of the least important criteria in this survey. All of these hoaxes will arrive by email from a known source, so it is questionable if the receiver of the email will try to find an authority source within the meme, other than the sender of the email. Nevertheless, for this research it is useful to check for the correlation between the authority of the hoax itself as well although it has to be kept in mind that the original authority (from the sender of the email) is not present in this survey.

The criterion of replication pressure or proselytism can logically be considered one of the most important ones in the spreading of the hoax. It is this criterion which will try to influence the receiver of the hoax to spread it. However, it is likely that a too high replication pressure will not lead to a higher replication rate but to a ridicule of the hoax. A hoax that consists only of 'please pass me on' messages may not be passed on due to the lack of content because people simply do not take it seriously (Hofstadter 52-53).

The Methodology Used in the Research

The Creation of the Survey

From the example hoaxes as provided by the virus information databases from several software houses, the format that most frequently appeared (i.e. that appeared in more than one virus information center) was chosen.

Each of the hoaxes was to be valued on the six criteria. The survey taker was asked to grade each of the criteria on a scale from one to five. There was no default value given to the criteria so a choice had to be made before the participant could continue with the following hoax.

When creating a survey one has to be aware of possible fluctuations in attention span of the survey taker. Questions which are asked earlier in the survey tend to get more attention than questions at the end of the survey.

To make up for this variation, three different versions of the survey were created. The six hoaxes that were selected for examination were divided into three groups of two hoaxes each. The criteria for these groups were that they could not contain two hoaxes from the same success-category of proliferation. The three proliferation groups were high, middle and low proliferation, based on the results of the searches as described in the section on the choice of the hoaxes. The first pair consisted of a low scoring hoax, California IBM and a hoax which scores average on fitness, Penpal Greetings. The second group is made up of one high scoring, Jdbgmgr.exe and one low scoring hoax, Buddylst. The third combination has an average scoring hoax, WTC Survivor combined with a high scoring hoax, Good Times.

These groups of hoaxes were placed in three different subsequent orders to ensure that each group was placed in the first part of the survey, the middle part of the survey and the end of each group in one of the surveys. Because of the way the pairs were organized a low scoring (California IBM), a middle scoring (WTC Survivor) and a high scoring hoax (Jdbgmgr.exe) were at the beginning of the survey and that a low scoring one (Buddylst), a hoax with an average fitness (Penpal Greetings) and a high scoring hoax (Good Times) were at the end of each group in one of the surveys.

Statistical Analysis

The results of the surveys are gathered per hoax and per question. The answers to the 36 different questions were imported into a spreadsheet program, Excel. By means of the data analysis module it was possible to make a statistical analysis and to compare the answers with the results from the search engine test in order to make a correlation between those two figures.

When correlation is measured it is important not only to pay attention to the positive correlation but to also keep the negative correlation in mind. A positive correlation will, in this research, indicate that the meme's fitness can be related to the criterion, a negative correlation will mean that the spreading degree of the meme is also dependent of this criterion but in an opposite way.

If the study would show that a high simplicity rating would correlate with a high fitness level of the meme, this would mean that the simpler the structure of a hoax is, the more chance this hoax has of being spread. If there would be a strong negative correlation this would mean that a hoax needs a complex language structure to have a higher spreading potential.

The Results of the Research

The hypothesis which this study would like to confirm is that the mentioned selection criteria will have a clear correlation with the success rate or fitness of the hoaxes. Criteria such as the ones that are mentioned in the research are thought to influence the spreading of the hoax.

The results of the survey as done by the expert in the field of memetics did not confirm the importance of all criteria. Instead three criteria stood out from the rest. Novelty, with a positive correlation of +0.5 and, even more so, benefit with a positive correlation of +0.8 seem to be the most important criteria. Simplicity has the lowest correlation with -0.6 to the fitness of the meme in both Search Engines and Usenet newsgroups and -0.53 when the correlation is measured with only the search engines. This means that if a hoax is perceived as being very simple, it will not reach a high degree of spreading.

The hypothesis of this thesis was that all selection criteria will have a clear correlation with the degree of fitness of the different hoaxes. After considering the results of both surveys, it is found that indeed there is a correlation, although it is not the positive correlation that was expected. Only two of the six criteria have a strong positive correlation with the success rate of the hoaxes. The other criteria are found to have a negative correlation, varying from very light (cf. proselytism) to a very strong negative correlation (e.g. simplicity). To validate these results and to make sure that they have not been caused because of a poor understanding of the selection criteria, the results were compared to those of an authority figure in the field of memetics. We see that this expert's scores have the same tendency and the same positive and negative correlations. Therefore the results can not be explained with this reasoning.

This research has shown that there is a wide variety in the ways the hoaxes use the criteria. Every hoax has found its own niche in the brains of several new hosts and on the Internet. Therefore all hoaxes here have already gone through the selection procedures that have been measured. The popularity of these hoaxes is different and the analysis of every single criterion that was measured in this survey has shed a new light on the ways in which hoaxes have adapted to the criteria in order to become as good or 'as fit' as possible.

The wide variance in techniques shows how exactly the hoaxes became successful. The importance of finding these techniques is arguably even greater than the failure to make a classification of the criteria. After all, it has become clear now that the criteria are very important but that the way in which the hoaxes use the different criteria has even more influence on their success rate.

For a more elaborate explication and analysis of each criterium, please download the full thesis.

Download Full Thesis PDF