Soapbox answer: Historians (at least good ones) almost never attribute a number of deaths to an ideology (because that is a logical fallacy taken to an extreme). Rather, historians attribute a death toll to an action, event, or combination thereof, and shite philosophers, politicians, pseudo-intellectuals pretending, political scientists and others who like to pretend to be historians lump a bunch of those events together, attribute them to communism, and declare communism to have the highest kill count ever.

Actual answer: Historians determine the death toll of events based on first and second hand accounts, evidence collected by archeologists or investigative bodies, and demographic information collected by governments. They then use any of a large number of statistical models to extrapolate an approximate death toll of the event. They can then take that death toll and examine its impact on events, societies, governments, etc. This is where a historian's role stops. Other individuals, or historians looking to make an ahistoric point, will then take these numbers for all of the events perpetrated by regimes which are currently called communist, add them together, and declare it communism's "death toll."

Necessary explanation: This is, of course, only stringently based in historic fact, because while the events may be real, the attribution of an event to an ideology is almost always fallacious. Attributing the anti-Jewish and anti-Ukrainian pogroms of Josef Stalin to communism is as idiotic as attributing the genocide of native Americans to monarchism. While the two coexisted, one was not the immediate cause of the other.

Specific question regarding Mao: the death tolls for Mao's Great Leap Forward are gathered or extrapolated from a few key sources (most of these can be found in some form online, but I am on my phone for the foreseeable future, so cannot provide them myself): demographic information reported by the CCP; information provided by government officials who defected from the CCP to other nations; the first hand accounts of victims and perpetrators who fled from China (for my favorite example of these, look for the book BORN RED - a truly horrifying look into history). To walk you through the process:

1. you gather demographic reports from the CCP and WHO that report the population of China changing from 175,000,000 to 180,000,000 with 30,000,000 births and 10,000,000 deaths over 10 years. Obviously, these numbers don't add up.

2. You look up testimony from a congressional hearing in the 1960's with a Chinese official who fled after the Cultural Revolution. In it, he admits that officials were encouraged to vastly underestimate the number of deaths in their province when people were starving. He claims he personally reported only 20% of all deaths. This would change the number of deaths to 50,000,000, with 30,000,000 births, which would mean the province should have lost 20,000,000 people when it actually gained 5,000,000.

3. You find an autobiography of a doctor during the Cultural Revolution who talks about how the stillborn births caused by mother's being malnourished were reported to the local officials as deaths, but not as births in order to enable them to falsify their numbers more easily. This explains the discrepancy earlier.

4. Based on all this, and other corroborating evidence which might help you develop a more exact estimate with better statistics, allows you to estimate the number of deaths in that province for that ten year period to be 50,000,000. This is just the raw number of deaths estimated for that ten year period. In my personal experience, this is often the (inaccurate) number attributed to 'deaths by communism' by the least factual and most political of actors.

To determine the number caused by the Great Leap Forward policy, specifically, we would generally take a few more steps, continuing with the example:

6. Determine that the number of deaths in the same province in a similar ten year period (for the purposes of our example, we are assuming we found two ten year periods with the same number and magnitude of huge natural disasters/wars that killed enough people to skew the numbers - this never fucking happens and requires another 40 steps to justify) came out to 30% of the total populace.

7. Given the numbers of the focus ten year period, 30% of 80,000,0000 would be 24,000,000 in a similar ten year period without the GLF policies. The focused on period showed 50,000,000 deaths, therefore it can be effectively argued that the GLF policies contributed to at least a portion of the 26,000,000 extra deaths when comparing the two ten year periods.

8. Because there are a wide number of variables we are assuming about, it is best to provide a probabilistic range, rather than a single number. Therefore, wherever we have made assumptions we should assume the worst, complete all the math, and determine the lower reach of our range, and then we should assume the best, complete the math, and use that as our upper limit. I haven't done this much math since I finished school, so let's just say we end up with between 16,000,000-30,000,000 deaths resultant from the Great Leap Forward.

Source: I did more-or-less this as part of an undergrad thesis. Except I had actual statistical models to plug numbers into. The example is vastly oversimplified for your reading pleasure. The real version had so much math my stats-major partner and I took turns having break downs over how much of the other's major we were being forced to learn to complete the models and papers.