Alan Schwartz
Department of Psychology
University of California, Berkeley

Many MUD researchers are interested in the prevalence of controversial behaviors such as gender-swapping (Bruckman, 1994; Reid, 1994). Unfortunately, few studies have been able to accurately measure the rates of gender-swapping or other such behaviors. Because players who admit to controversial behaviors may receive adverse reactions from other players, players are motivated to conceal these behaviors. If players could be sure of anonymity, they might provide truthful responses, but email surveying places the power of anonymity in the hands of the researcher, who must actively strip email addresses from the responses. On-line surveys are often vulnerable to breaches of anonymity by MUD administrators.


Probabilitistic survey questions offer one means to assure anonymity while providing estimates of the rates of behaviors. The basic technique asks respondents to follow these steps:

  1. Flip a coin. If it lands heads, you are in the "coin" group. If it lands tails, you are in the "question" group.
  2. If you are in the coin group, flip a coin again.
    • If it lands heads, respond "yes".
    • If it lands tails, respond "no".
  3. If you are in the question group, answer the controversial question honestly, e.g. "Are you of the same gender as your character?"


After N players have responded as directed above, it's possible to estimate the rate of behavior as follows:

  1. Assume that of the N players, Y of them responded "yes".
  2. Of the N players, on average N/2 of them will be in the coin group. Of this group, half will respond "yes" and half "no". That is, N/4 players respond "yes" by chance, and N/4 players respond "no" by chance.
  3. Therefore, Y-N/4 players responded "yes" to the question itself, and (N-Y)-N/4 players responded "no" to the question itself. The percent of the sample performing the behavior is (Y-N/4)/(N/2)
For example, if 35 players out of a sample of 100 respond "yes", on average, 25 of the "yes" responses (and 25 of the "no" responses) will be from the coin group. Discarding these responses leaves 10 true "yes" responses and 40 true "no" responses, or a 20% prevalence.


The advantages of probabilistic questions become apparent when you consider what the investigator (or a MUD administrator) knows and doesn't know about an individual response. The investigator may know the character who made the response, so participation in the question is not anonymous. But whether the response is "yes" or "no", the investigator does not know whether any particular response was generated randomly or by actually responding to the question.

To maximize the value of these questions, the players themselves must understand the process and how it affords them anonymity. In particular, it should be made clear that individual responses can not be identified as random or nonrandom, and that they themselves determine whether their responses are random by a totally private coin flip which no one else can observe.


Of course, the advantages of these questions do not come for free. Most serious of the limitations is that probabilistic questions can do little more than estimate rates of behavior. While it is possible to construct more complex procedures which yield contingent probabilities (e.g. the rate of gender swapping among RL males vs. RL females), these procedures rapidly become too complex to explain to subjects. Hypotheses about correlations between individual variables are nearly impossible to assess in this way.

A second limitation is that subjects can still respond dishonestly and produce biased estimations. This is a general problem in surveying, however, and seems unlikely to be worsened by the probabilistic format, particularly if subjects are made clearly aware of the ways in which their anonymity is protected.


Bruckman, Amy. (1994). Gender Swapping on the Internet. Available: (19 May 1996).

Reid, Elizabeth. (1994). Cultural Formations in Text-Based Virtual Realities. Master's Thesis. Available: (19 May 1996)