Lizardman's Constant is the informal scientific theory that ~5% of humans will respond to surveys as if they are batshit insane.
The term was popularized, and perhaps coined, by Scott Alexander1, who noted that in a 2013 telephone poll which asked "Do you believe that shape-shifting reptilian people control our world by taking on human form and gaining political power to manipulate our society, or not?", 4% of Americans respond that they did indeed believe this. (Additionally, another 7% were unsure.)
One theory for this sort of response is that people are not trying to respond with a true answer, but with a truthy answer -- they are trying to find an answer that reflects the depth of their feelings, rather than boorishly reflecting the mundane world. This explains the results from the same poll finding that 13% of Americans believed that Obama was the Anti-Christ. Presumably, the lizardman people were simply expressing a strong dislike for politicians, and the Anti-Christ people a negative opinion on Obama, with neither group intending to make extreme herpetological or religious claims.
There is perhaps a stronger theory: humans are perverse, and will lie for fun. This theory recently gained a lot of attention when a research paper2 was published highlighting the responses given by thousands of teenagers (grades 9–12; about 14-18 years of age) on surveys targeting topics like sexual identity, gender identity, alcohol use, and physical disability. They found that "mischievous responders" skewed a number of results. They focused primarily on the ability of small numbers of extreme responders to skew results3 rather than the absolute percentages, which limits the data regarding lizardman's constant. However, they found that in one survey ~0.6% of students lied on the question as to whether or not they were adopted, while on another poll when they asked the students if they had responded truthfully to all questions, 12% admitted that they had not. One factor here is that many false statements cannot easily be disproved; while it can be determined that a person is not adopted, it is much harder to prove that they are not secretly drinking. However, they also noted that and about 2% of respondents provided three or more low-frequency responses, indicating that they might be "mischievous".
There are also other possible causes for lizardmaning; inattention and answering at random, revenge against a telephone pollster calling during dinner, manic glomming on to interesting oddities, or honestly believing in lizardmen. It is also apparent that lizardman's constant isn't constant; it appears to be highest in internet polls, midrange on telephone polls, and lowest in physical paper and pencil polls.
Regardless of the cause, some surveys opt to contain 'control questions', e.g. "My favorite poet is Raymond Kertezc", which gives the researchers warning that this respondent is inattentive, answering at random, or lying. However, it appears this is mostly used to disqualify these people, and I have not found specific numbers on the average percent of respondents claiming to be Raymond Kertezc fans.
There is also the closely related crazification factor (AKA the Keyes constant), which states that ~20-30% of the population are nutcases; in practice, this is the highly informal statement that you can find a large percent of people who will agree with popular fringe beliefs (e.g., the Flat Earth Society), rather than the more formal statement that you can find about 5% of respondents agreeing with anything.
1. The original blog post. It is well worth reading.
2. Robinson-Cimpian, J. P. (2014). Inaccurate Estimation of Disparities Due to Mischievous Responders: Several Suggestions to Assess Conclusions. Educational Researcher, 43(4), 171–185. link.
3. e.g., if there are very few children who are adopted, and a small number of children who falsely report that they are adopted, if the false reporters also report that they are heavy drinkers, this can falsely suggest that adoptees are early heavy drinkers. This effect was certainly apparent in self-reported adoptees (and "adoptees"), but was much greater in self reported amputees, as there were significantly fewer true cases of amputees (there were 2) and lots of the jokers (251 of them), which completely swamped the sample.