Western sitting-style toilets are presumably intended to be used in a seated position, (McClelland & Ward, 1982) except when used by males for exclusively urination. A sign on a bathroom in my college dorm illustrates this expectation.

Sign on a door with symbols for both sex and a stylized person sitting on a stylized toilet

But the word on the street is that people often adopt a hovering posture instead of the seated posture in order to cope with unsanitary toilets. To start, several threads on Yelp contain discussion and complaint about toilet cleanliness and preferred postures. Also, one person admitted to PostSecret that he or she hadn't sat on a toilet for the four years.

I haven't sat on a toilet seat in over 4 years.

(This sparked further discussion on PostSecret forums.)

There's even a website that reviews toilets and tells you whether to Sit or Squat.

Greed (Greed, 1995) and Moore (Moore & al, 1991) anecdotally found that young ladies are often taught that hovering is proper and that sitting is unclean. Several WikiHow articles provide examples of this phenomenon.

Formal quantitative studies have found hovering to be common among Taiwanese people (Cai & You, 1998) and British gynaecological outpatients (Moore & al, 1991),

Toilets get dirty, so people hover.

Many of the referenced forum discussions explain that hovering, rather than sitting, creates more of a mess. Presumably, this is because hovering puts the relevant body parts in a less stable position that is higher above the toilet, making it harder to aim.

If these anecdotes about the causes and effects of hovering are correct, then the clean toilet is an unstable equilibrium of a toilet's sanitary state; toilets get dirty, so people hover, so the toilets get dirty, so people hover.

Cycle of hovering and toilet cleanliness

In other contexts, this cycle is sometimes termed the "broken windows theory", or the feeling of "maintenance" of a space.

Effects of privacy on posture choice

I focused on this broken windows theory in the current paper, but I also thought about other impacts on posture. Anecdotes suggest that public toilets tend to be dirtier than private toilets, or people or at least more concerned with sanitation at public toilets than at private toilets (Cai & You, 1998), so we might expect that a more public toilet would effect posture similarly to how a more dirty toilet would.

On the other hand, privacy might have a secular impact on posture choice. For example, social norms might have more of an impact in public toilets than in private toilets. A lady once told me that she thought people would be more likely to hover at public toilets because hovering is considered more proper than sitting; since people can see others' feet and lower legs under the bathroom stalls in public restrooms, people might be more likely to follow this convention of hovering in public than in private.


Understanding the mechanisms of toilet cleanliness can help us design and maintain efficient toilets that people want to use, so it would be nice to see whether this the broken windows theory explains toilet sanitation issues.

I sought to establish whether this cycle of hovering and toilet dirt exists by looking at the postures that university students say they'd use in toilet-use scenarios of differing privacy and cleanliness. I was mainly concerned with how cleanliness influenced the decision of what posture to use, but I tried to separate effects of privacy from effects of cleanliness.


Western toilets typically consist of a bowl and a seat with a large hole in the middle. The seat typically can lie on the bowl or be lifted above the bowl. I consider four categories of postures in which this contraption can be used.


Sitting-style toilets are designed to be used in a seated posture (McClelland & Ward, 1982), which we define as one where the buttocks or thighs touch the toilet.


Hovering, also called squatting or semi-squatting, describes postures where the feet are on the floor but the buttocks and thighs are above the toilet. This is accomplished through knee extension and trunk flexion. (Greed, 2003)


In squatting, the thighs touch the legs and the trunk touches or nearly touches the thighs. In order to accomplish this on a sitting-style toilet, one puts his or her feet on the toilet seat. This is probably much less common in countries where sitting toilets are the norm.

Confusingly, term "squatting" is also often used to refer to what I term "hovering".


Males commonly stand at toilets when they are urinating. In this posture, the person faces the toilet and stands in front of the toilet.


In May 2011, I sent a questionnaire to randomly selected Cornell University students (graduate and undergraduate). The questionnaire asked more questions than I presently present, but I'll just explain the relevant questions.


I asked each participant for his or her sex. Participants were required to choose the more appropriate of "male" or "female". I also asked each participant what posture he or she would use in 12 different scenarios. For each scenario, the participant was required to respond with one of these postures:

The 12 scenarios came from a three variables in a 2 × 2 × 3 layout. The variables were

For example, one question was

If you are urinating at a very dirty private toilet, which posture are you most likely to use?

The "unspecified" cleanliness level meant that cleanliness level was not mentioned; for example, another question was

If you are urinating at a public toilet, which posture are you most likely to use?

Validity testing

I pilot-tested the questionnaire on nine people. For each of these pilot test participants, I asked the participant to complete the questionnaire, and then I interviewed the participants to see how they had interpreted each of the questions. I made some changes based on this pilot study, but all of the students responded the same in person as they had on the questionnaire. I found the participants in the psychology department's participant pool.

Main sample

I selected random students from the school's public electronic directory. (It's the same directory as the one from this study on middle names at Cornell University.

I sent emails to 390 students. The emails asked the participants to complete the questionnaire online in exchange for a one-in-three chance of winning $10. For students who did not respond, I sent up to two reminder emails.

Reliability testing

A week after the survey emails were sent, I selected the 20 people who had responded first. I sent them a very similar email asking that they take the questionnaire again in order to test it's reliability. These participants received the same compensation as they had received in the main survey phase—another one-in-three chance at winning $10, even if they had won before.


Of the 390 students to which questionnaires were sent, 173 completed the questionnaire. There were 103 females and 70 males.

Popularity of posture by situation

Within each of the 12 situations, the most popular posture was normally sitting. We can see more in these plots. Each circular thingy represents the posture choices for a question about a particular task and privacy with no mention of cleanliness (like "If you are urinating at a private toilet, which posture are you most likely to use?"). The top row is the male responses, and the bottom row is the female responses.

Polar plots of posture choice (sit, stand, hover, other) by scenario

The table below presents the popularity of the two most popular postures for each situation. By "popularity", I mean the proportion of students who would say they use the posture if I asked.

sex task privacy cleanliness hover other sit stand
male defecate private unspecified 1 (1%) 3 (4%) 66 (94%) 0 (0%)
male defecate private clean 2 (3%) 0 (0%) 64 (91%) 4 (6%)
male defecate private dirty 38 (54%) 3 (4%) 26 (37%) 3 (4%)
male defecate public unspecified 6 (9%) 4 (6%) 60 (86%) 0 (0%)
male defecate public clean 2 (3%) 1 (1%) 64 (91%) 3 (4%)
male defecate public dirty 43 (61%) 5 (7%) 19 (27%) 3 (4%)
male urinate private unspecified 1 (1%) 4 (6%) 13 (19%) 52 (74%)
male urinate private clean 1 (1%) 0 (0%) 12 (17%) 57 (81%)
male urinate private dirty 5 (7%) 0 (0%) 1 (1%) 64 (91%)
male urinate public unspecified 4 (6%) 4 (6%) 2 (3%) 60 (86%)
male urinate public clean 2 (3%) 0 (0%) 6 (9%) 62 (89%)
male urinate public dirty 4 (6%) 0 (0%) 1 (1%) 65 (93%)
female defecate private unspecified 1 (1%) 3 (3%) 97 (94%) 2 (2%)
female defecate private clean 6 (6%) 2 (2%) 94 (91%) 1 (1%)
female defecate private dirty 59 (57%) 14 (14%) 27 (26%) 3 (3%)
female defecate public unspecified 18 (17%) 3 (3%) 81 (79%) 1 (1%)
female defecate public clean 29 (28%) 1 (1%) 71 (69%) 2 (2%)
female defecate public dirty 59 (57%) 18 (17%) 22 (21%) 4 (4%)
female urinate private unspecified 6 (6%) 2 (2%) 93 (90%) 2 (2%)
female urinate private clean 10 (10%) 0 (0%) 91 (88%) 2 (2%)
female urinate private dirty 81 (79%) 5 (5%) 11 (11%) 6 (6%)
female urinate public unspecified 55 (53%) 2 (2%) 44 (43%) 2 (2%)
female urinate public clean 45 (44%) 0 (0%) 55 (53%) 3 (3%)
female urinate public dirty 81 (79%) 8 (8%) 7 (7%) 7 (7%)

The popularity of sitting makes sense given that they are called "sitting toilets". Notable exceptions are when men are urinating (They stand) and when women are urinating at dirty or public toilets.

The "other" option was especially popular for dirty toilet situations. I suspect that participants chose this to indicate that they would never ever under any circumstances ever use such a toilet.

Reducing dimensionality

I converted the categorical "posture" variable into a binary "dirty posture" variable. I considered "hover" and "other" to be dirty, and I considered "sit" to be clean. I considered "stand" to be dirty unless it was marked by a male for a scenario involving urination. (Standing is actually pretty dirty, and some males choose to sit, but I see standing as the conventional, baseline posture for these scenarios.) Nobody had marked "squat", so that didn't need to be converted.

That results in this table. It shows the numbers and proportions of participants who said they used dirty postures in response to each of the questions.

sex privacy task cleanliness Dirty posture
male private defecate unspecified 4 (6%)
male private defecate clean 2 (3%)
male private defecate dirty 41 (59%)
male private urinate unspecified 5 (7%)
male private urinate clean 1 (1%)
male private urinate dirty 5 (7%)
male public defecate unspecified 10 (14%)
male public defecate clean 3 (4%)
male public defecate dirty 48 (69%)
male public urinate unspecified 8 (11%)
male public urinate clean 2 (3%)
male public urinate dirty 4 (6%)
female private defecate unspecified 6 (6%)
female private defecate clean 9 (9%)
female private defecate dirty 76 (74%)
female private urinate unspecified 10 (10%)
female private urinate clean 12 (12%)
female private urinate dirty 92 (89%)
female public defecate unspecified 22 (21%)
female public defecate clean 32 (31%)
female public defecate dirty 81 (79%)
female public urinate unspecified 59 (57%)
female public urinate clean 48 (47%)
female public urinate dirty 96 (93%)

Does dirt promote dirt?

The above table about the use of dirty postures versus clean postures gives us an answer this question. If dirty toilets promote dirty postures, we should expect the rate of dirty postures to increase as toilets get dirtier. The difference is quite strong, so I could use a low-powered but easily-interpreted method to check that the difference was significant.

I estimated a 95% confidence interval based on a normal approximation of the binomial distribution. I conservatively assumed that the proportion was 0.5 when computing the standard error. Also conservatively, I only computed it for the male sub-sample (n = 70 rather than n = 173).

SE ( p ) = ( 1 - ) n - 1 = 0.5 2 69

Then multiply it by z = 2 in each direction to get 12%, so we can be quite confident that these percentages figures are within 12% of the actual. Well actually, because of the confidence level I chose, we should expect the proportion for one of the 24 above-tabulated scenarios to fall outside confidence interval.

This tells us that Cornell students in general are more likely to use dirty postures as toilets get dirtier.

Do public toilets affect posture choice for reasons other than dirt?

I checked simply whether posture was the same between the corresponding public and private conditions for each participant. This is six comparisons per participant, for each of the cleanliness × task combinations.

There were instances where posture was different, so the quick answer is "yes". But how often were they different? To avoid talking in double-negatives, I actually checked how often the posture was the same. Here are the raw proportions.

Task Cleanliness Male Female
Urinate Clean 99% 65%
Unspecified 96% 50%
Dirty 96% 96%
Defecate Clean 99% 78%
Unspecified 91% 83%
Dirty 87% 93%

I also plotted them, separating by sex. The error bars are 12% tall in each direction; the reasoning for this is the same as in the earlier table.

Polar plots of posture dirtiness (clean versus dirty) by scenario

In most situations, the figures are quite close to 100%, but a couple are quite relatively low. In particular, the figure is 50% for females in the unspecified-urinate condition; this is to say that half of females who responded to the questionnaire said they would use a different posture when "urinating at a private toilet" than when "urinating at a public toilet".

I was originally not so concerned with the difference between sexes, but it seems interesting now, so let's run some fancy significance test. I fit a multivariate analysis of variance. Sex was the sole independent variable, and the correspondence of postures between the six sets of corresponding public and private scenarios were the response variables. I haven't figured out a straightforward way of writing that, so here are three rows from the response variable matrix.

unspecified&defecate unspecified&urinate clean&defecate clean&urinate dirty&defecate dirty&urinate

Cells say "TRUE" if posture was the same in the corresponding public and private scenarios, and they say "FALSE" if it wasn't.

We saw in the graph that sex has the largest impact on the difference between the public and private locations for the unspecified_urinate, clean_urinate and clean_defecate scenarios. We see the same thing in the MANOVA coefficients.

                       person$sex Residuals
unspecified_defecate         0.33     20.34
unspecified_urinate         8.526    28.619
clean_defecate              1.821    18.850
clean_urinate               4.684    24.403
dirty_defecate              0.153    14.367
dirty_urinate               0.001     6.716
Deg. of Freedom                 1       171

Residual standard error: 0.3449 0.4091 0.332 0.3778 0.2899 0.1982 
Estimated effects may be unbalanced

All of the standard four hypothesis tests suggest that sex has a significant difference on the difference between public and private scenarios.

            Df           Pillai approx F num Df den Df  Pr(>F)    
person$sex   1            0.268     10.1      6    166 1.6e-09 ***
Residuals  171

            Df            Wilks approx F num Df den Df  Pr(>F)    
person$sex   1            0.732     10.1      6    166 1.6e-09 ***
Residuals  171

            Df Hotelling-Lawley approx F num Df den Df  Pr(>F)    
person$sex   1            0.365     10.1      6    166 1.6e-09 ***
Residuals  171

            Df              Roy approx F num Df den Df  Pr(>F)    
person$sex   1            0.365     10.1      6    166 1.6e-09 ***
Residuals  171

I ran univariate ANOVAs for the six scenarios.

             Df Sum Sq Mean Sq F value Pr(>F)  
person$sex    1   0.33   0.330    2.78  0.097 .
Residuals   171  20.34   0.119

             Df Sum Sq Mean Sq F value  Pr(>F)    
person$sex    1   8.53    8.53    50.9 2.6e-11 ***
Residuals   171  28.62    0.17

             Df Sum Sq Mean Sq F value  Pr(>F)    
person$sex    1   1.82    1.82    16.5 7.3e-05 ***
Residuals   171  18.85    0.11

             Df Sum Sq Mean Sq F value  Pr(>F)    
person$sex    1   4.68    4.68    32.8 4.5e-08 ***
Residuals   171  24.40    0.14

             Df Sum Sq Mean Sq F value Pr(>F)
person$sex    1   0.15   0.153    1.82   0.18
Residuals   171  14.37   0.084

             Df Sum Sq Mean Sq F value Pr(>F)
person$sex    1   0.00  0.0007    0.02    0.9
Residuals   171   6.72  0.0393

The MANOVA doesn't tell us anything we didn't already figure out from the graph; all of the MANOVA significance tests find a difference by sex in joint rate of correspondence across scenarios, and univariate ANOVAs on the six respective variables indicate significant differences for the unspecified_urinate, clean_defecate and clean_urinate conditions These are the three that are obviously different based on the graph.

Using MANOVA actually isn't quite appropriate because it treats the response variables as continuous rather than being only one or zero. It would be more appropriate to use generalized estimating equations, but I don't really know how those work and don't care enough right now. Because of this inappropriateness, the residuals are negatively skewed,

I wouldn't worry too much about the validity of the MANOVA, but you should keep in mind that I might not really be distinguishing between privacy and cleanliness. Considering the nature of the data collection, it is hard to tell whether this analysis really does disentangle toilet privacy and toilet cleanliness. Participants were asked about "clean private toilet[s]" and "clean public toilet[s]", and they could have interpret the word "clean" differently in the two contexts; a clean public toilet might just be clean relative an ordinary public toilet. If participants saw clean private toilets as cleaner than clean public toilets, this result about the difference between public and private scenarios doesn't separate the effect of cleanliness from the other effects of the privacy of the space.

Anyway, it appears that a toilet being dirty doesn't affect posture much after I control for everything else. The situations where it does effect posture are when females are in the less unpleasant situations of urinating at non-dirty toilets.

Characterizing people

The analysis thus far has looked for changes by scenario, across people. For example, I found that more people use dirty postures (mostly hover) when toilets are dirty. I haven't yet looked at all at how particular people change their postures as scenarios change. Said differently, I haven't looked at whether the people who hover when toilets are clean are the same as the people who hover when toilets are dirty.

Plots by person

First, I made a separate plot for each participant to characterize that participant's posture choice. I ordered the scenarios by likelihood to use a posture other than the standard sitting (or standing for males urinating) and made separate files for males and females.

Flipping through all of those is sort of interesting, but I would like to condense the trends into something more easily digested.

It seems like we can vaguely order the various scenarios in a particular way such that we can usually split the scenarios at two along that ordering and find that a person uses dirty postures at most scenarios on one side and uses clean postures at most scenarios on the other side. More on that later.


I used hierarchical clustering to group the participants based on their responses to the twelve questions. I used the Ward method with euclidean distances. The grouping looks like this.


We can cut off this dendrogram at any level and look at the average (really the centroid) value within each cluster. If we cut it before the first split and have one "cluster", the centroid will just be the the values in the table in the "Do public toilets affect posture choice for reasons other than dirt?" section, except not grouped by sex. (I clustered based on question responses only, not based on sex.)

I cut off the dendrogram at successively lower levels until it got boring, annotating the clusters as I went. Here's the first cut.

Dendrogram with boxes around two clusters

And here are the centroids. (I didn't cluster based on sex, but I included sex in the table below.)

Variable/Scenario Cluster 1 Cluster 2
male 0.71739 0.04938
unspecified_defecate_private 0.03261 0.08642
unspecified_defecate_public 0.07609 0.30864
unspecified_urinate_private 0.04348 0.13580
unspecified_urinate_public 0.14130 0.66667
clean_defecate_private 0.01087 0.12346
clean_defecate_public 0.03261 0.39506
clean_urinate_private 0.00000 0.16049
clean_urinate_public 0.03261 0.58025
dirty_defecate_private 0.41304 0.97531
dirty_defecate_public 0.53261 0.98765
dirty_urinate_private 0.18478 0.98765
dirty_urinate_public 0.21739 0.98765

After studying the centroids, I came up with these groupings.

Dendrogram with boxes around two clusters, one labeled "male" and the other labeled "female"

Here it is at six clusters, grouped.

Dendrogram with boxes around six clusters

Variable/Scenario Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6
male 0.77 0.73 0.07 1 0 0
unspecified_defecate_private 0.14 0.00 0.12 0 0 0
unspecified_defecate_public 0.31 0.00 0.44 0 0 0
unspecified_urinate_private 0.18 0.00 0.19 0 0 0
unspecified_urinate_public 0.23 0.00 0.95 0 0.57 0
clean_defecate_private 0.05 0.00 0.18 0 0 0
clean_defecate_public 0.13 0.00 0.56 0 0 0
clean_urinate_private 0.00 0.00 0.23 0 0 0
clean_urinate_public 0.00 0.00 0.82 0 0.21 0
dirty_defecate_private 0.36 0.00 0.96 1 0 1
dirty_defecate_public 0.86 0.00 0.98 1 0 1
dirty_urinate_private 0.18 0.00 0.98 0 0.93 1
dirty_urinate_public 0.27 0.00 0.98 0 1 1

I stopped at 17.

Dendrogram with boxes around 17 clusters

The above groupings help us understand the questionnaire responses by person, but they don't tell us how well the groupings might apply to people outside the sample. To check how robust these groups are, I used approximately unbiased p-values, computed by multiscale bootstrap resampling in pvclust. I only vaguely know how this works: Resamples of varied sizes are taken from my 173-person sample, and the stability of the results from these resamples is computed.

It found most of those last 17 clusters to be significant, (That is, we wouldn't have seen them unless they're real.) with approximately unbiased p-values greater than 0.95. Unexcitingly, the p-values were zero for most of the higher clusters. Though I don't really know how this technique works, I suspect that this result is somewhat explained by my use of cluster analysis to describe binary (yes/no) data. Anyway, I thus don't have a particularly good confidence figure for the clustering.

Counts of dirty postures by person

Having looked so much at the within-person relationships between different the scenarios, what general things can we say about the way people make decisions?

The plots and clustering generally suggest that a particular person is more likely to use a dirty posture as the toilet changes from clean to dirty and from private to public. Also, a particular female is more likely to use a dirty posture when she urinates at than when she defecates at an equivalent toilet.

Said differently,

Recall that each participant marked what posture he or she would most likely use in each of twelve scenarios and that the postures were grouped into clean and dirty postures. Since we are able to narrow down the marginal reasons why a person would switch postures, the following plot is informative.

How many participants marked how many scenarios?

The x-axis is the number of questions for which dirty postures were marked, and the y-axis is the proportion of participants. There is a separate curve for each sex, and the area under each curve is 100%. Because males rarely use dirty postures when urinating, it might make sense to think of the male curve as ending at six questions rather than at twelve.

For both sexes, most people rarely use dirty postures. On the other hand, few people always use clean postures. Moreover, based on our ranking of the different scenarios, it seems that about half of people of either sex will use dirty postures in the dirty scenarios and only in the dirty scenarios.

General trends

As I just said, a particular person is more likely to use a dirty posture as the toilet changes from clean to dirty and from private to public. Also, a particular female person is more likely to use a dirty posture when she urinates at than when she defecates at an equivalent toilet.

These conclusions match the conclusions based on the aggregated rates of use of the various postures, but these are at the level of participants rather than the level of the group of participants. Previously, we only could say that more people are likely to hover as toilets get dirtier; now, it appears that particular individuals are more likely to hover as toilets get dirtier. (That might sound like common sense.)

Reliability testing

I sent a second questionnaire to 20 people, and 16 responded. For each person's response to each question, I checked whether the responses from the two rounds matched each other.

task cleanliness privacy Same Different
defecate clean private 16 0
defecate clean public 15 1
defecate dirty private 14 2
defecate dirty public 16 0
defecate unspecified private 15 1
defecate unspecified public 13 3
urinate clean private 15 1
urinate clean public 15 1
urinate dirty private 14 2
urinate dirty public 14 2
urinate unspecified private 16 0
urinate unspecified public 16 0

For the least consistent question, 13 out of the 16 people (81%) had the same response for both rounds. So the responses were quite consistent between the test and retest questionnaires.


Broken windows

At the beginning of this paper, I presented the following hypothesis.

Toilets get dirty, so people hover, so toilets get dirty.

The present study supports this hypothesis; people were more likely to say that they'd hover or use a different dirtying posture at dirty or public toilets than at clean or private toilets. This was true both when we looked at the group as a whole, not considering changes within a person, and when we looked at the level of individual persons.

For a more complete verification of this hypothesis, another study could observe how posture choice affects the dirtying of the toilet and see whether my categorization of postures into clean postures and dirty postures is reasonable.

Do public toilets promote hovering for reasons other than dirtiness?

The study also gives us an idea of how people might decide whether to hover when using a public toilet or a dirty toilet.

Dirtier toilets and more public toilets encourage hovering and other dirty postures. Toilet cleanliness and privacy might be correlated with each other, (I didn't address that in the present study.) but they also might have effects separate from each other. The data seem to suggest that toilet privacy affects posture for reasons other than toilet cleanliness, but the manner in which the questionnaire questions were asked makes it difficult to claim whether I have truly isolated the effect of privacy.

Men and women

Moreover, the study shows how the cleanliness of a bathroom might affect different groups differently.

When using a clean or private toilet ("pleasant toilet" henceforth), people are more likely to use a clean posture. More specifically, men and women are both unlikely to hover while defecating, presumably because it is difficult.

For use of an dirty or public ("unpleasant") toilet, the differences between males and females become relevant. For defecation, both sexes generally choose to sit. For urination, men generally choose to stand, and women generally choose to hover. This switch demonstrates a compromise between sanitary comfort and musculoskeletal comfort.

People would prefer not to have to touch the unpleasant toilet. Men and women pee in different ways and thus accomplish this in different ways; men stand, and women hover. Standing is presumably much easier than hovering, so most men stand during urination, while only about half of women hover. (The rest mostly sits.) Men and women defecate more similarly, so when they switch to defecation, the rates of clean and dirty postures are closer. The rates are still a bit higher for women, though; it might be that women have more practice in hovering and thus are more likely to hover at an unpleasant toilet.

The current study compared males and females, but loads of other things that could affect how people use a toilet. These things include anthropometry, fatigue and disabilities.

Toilet design

Toilets are called "sitting" toilets, but lots of people hover, even in the most desirable of conditions. Something is wrong here; the design and conventions of toilets do not match the way that toilets are used.

If people want to hover, maybe we should let them hover. Would adding bars to stalls help people hover? And would that, in turn, make people happier?

Maybe we can make toilets fit a wider range of people. Men have the luxury of calmly standing while urinating, whereas women have to either hover or bear touching the seat. Could toilets be designed in some way that matched the female anatomy? Clara Greed (Greed, 2003) might give us ideas. More generally, can we design healthful toilets that still align with our cultural expectations? Alexander Kira (Kira, 1976) has ideas on that.

The insistence on hovering over a public toilet can also be seen as a silly obsession with cleanliness. Sitting on the public toilet is probably perfectly safe, and hovering is known to have undesirable health outcomes, like taking longer to pee (Moore & al, 1991). Can we just get over our concern for such cleanliness?

As an aside, we should consider squatting. Among other benefits, it might make peeing faster. (Amjadi & al, 2006; Rane & Corstiaans, 2008) On the other hand, it might not. (Unsal & Cimentepe, 2004)

Keeping toilets clean

Public bathrooms are notoriously dirty. (They even scare children. (Vernon & al, 2003; Lundblad & Hellström, 2005) The present study provides insights into how we can efficiently reduce this dirt. The current study has related the broken windows theory to the cleanliness of bathrooms. When people use dirty bathrooms, they tend to use postures that make toilets dirtier. Thus, clean bathrooms stay clean, but dirty bathrooms get dirtier.

Perhaps we can schedule bathroom cleanings better with this knowledge. Rather than cleaning bathrooms once a day or waiting until they get messy, it may make sense to quickly tidy them up every hour or so; wipe the toilets and the sinks, pick trash up from the ground, and flush any toilets that haven't been flushed. This may stop toilets from getting particularly dirty, allowing toilets to stay reasonably clean throughout the day without major cleaning.

We can also install signs in the bathroom that encourage people to clean up minor messes. When I was in my senior year of college, the cleanliness of my dorm's bathrooms became a topic of epic controversy. Around that time, I observed a couple signs that requested that men avoid getting urine on the toilet seats and that they wipe the toilet in case they did get urine on the toilet.

Letter written on notebook paper and posted inside a bathroom stall encouraging gentlement to be neat

Annotated drawing of a toilet posted inside a bathroom stall encouraging gentlement to be neat

The above notices are phrased in a way that is somewhat hostile towards males; they might be more effective if they requested that less unpleasant unsanitary conditions be avoided. Based on the results from the present study, I suspect that the prevention of minor messes that people don't complain about would prevent the major messes that people complain about.

A sign at General Assembly ungrammatically makes this sort of request.

Sign requesting people to flush the toilets

A bathroom in the Columbia University School of Journalism contains a couple of similarly small requests.

Please don't throw paper towels in the urinals!

This is a common bathroom, and the Journalism School's hard working staff must maintain it. Please clean up when you are done, and do not leave toilet paper on the floor.

By asking people to keep the bathrooms clean in small ways, we may delay the onset of this vicious hovering/dirt cycle.


54% of women hover when urinating in public toilets.

The estimated prevalences of various postures of toilet use make for fun dinner conversation.

Helpful tools

I ran the questionnaire on Qualtrics. Qualtrics gave me a verbose and denormalized spreadsheet, so I I used SQLite and dumptruck to import the data, then I studied everything in R, using ProjectTemplate, hclust, cluster, pvclust, manova, and all of Hadley's libraries. The paper is mostly compiled with knitr and R Markdown, and I manually edited the resulting markdown a bit.

Proprietary references