What are the two major misconceptions of data analysis

At ordinary times, whether the user research, product operation or competitive analysis, the analysis of the data, if there are some differences in the project who can not convince who, many times is also to take data to speak, visible in the development of products, statistics and analysis of the data is very important. Everyone said that the data is objective, but in fact, the data by the background environment, statisticians, statistical methods, analysts, such as the impact of multiple factors, so that we are in statistics and analysis often fall into the wrong, not the right answer. Let's say a few two common misconceptions about data.

Mistake one: A type of data as all data results in error analysis

Let's start with a little story: in World War II, the British Air Force wanted to increase the armor thickness of the aircraft, but if all the armor thickening would reduce flexibility, the final decision was to increase the armor at the most attacking sites. Later, the staff after the statistics of the bullet plane, found that most of the aircraft wing bullet holes more, so decided to increase the armor thickness of the wing. Later an expert said: "But the plane that was shot in the nose did not fly back."

This story is supposed to be an analysis of all airplanes, but the statistical sample does not contain the damaged aircraft, so the conclusion is based on part of the data, or based on the same characteristics (injury) of a class of data inferred, and can not represent all types of data, so the results are likely to be wrong.

Another example: a while ago, in order to analyze Renren, want to see how people are now user access to the state, and then selected PV as an observation index, through Alexa to see Renren PV in the past year has shown a significant decline, which also confirms my expectations, so on this as an argument for analysis. But later found that Alexa only statistics through the Web Access, and the user's mobile end of the login is not in the statistical range! The popularity of smartphones in the past two years, mobile-side logins are also very common, missing this part of the data means that the previous statistics are basically meaningless, Because the decline in web-side access is likely to be a decrease in user access to Renren, and possibly from a PC-side migration to the mobile side, this statistic cannot be used as an argument.

As you can see from the example above, I've only counted the web-side access, and I think it's all about Renren, and ignoring the mobile side, which introduces the wrong results. Another problem is that because I already have expectations (a decline in renren), I'm looking for the relevant arguments for this conclusion, and when it comes to finding arguments that match my conclusions, it's easy to choose data that is good for me without making more judgments, which is a common problem for data statisticians.

Replacing all of the data with a certain type of data would mislead us into making false judgments, and we must pay attention to this in statistics. This need to be aware of, in the statistics, analysis of data must always think about whether there are other circumstances, there is no we do not think of data types, the data is not representative of all types, try to stand in a higher perspective to interpret the data, rather than get the data immediately after the blind analysis. On the other hand, you need to accumulate knowledge, for example, you know how the Alexa statistics, it is very easy to think of also consider the mobile side of the situation. The accumulation of knowledge helps us to make accurate judgments, these knowledge and experience are derived from reading or practice, usually do more, slowly accumulate, time will naturally see more comprehensive.

Myth Two: The stark events let us exaggerate the accidental factors

Bright events are more likely to occupy our sights, allowing us to overestimate the probability of an event happening.

For example, seen from the annual statistics, a fund in the past two years, the yield reached 100%, there are such stars and so on, people will scramble to buy the fund, but also make people think that the purchase of funds can make money. In fact, there are few funds that can keep this rate of return year-round, in the past two years, the first five of the fund is likely to be in the bottom five years, and most of the fund can not run to win the market, but people still think that the purchase fund is really very lucrative, the same year XXX two years benefit 100%. A two-year gain of 100% was a fluke, but it was long in the minds of the people because the events were too stark.

There are a lot of things like that. Like the Fuji-N Hop, everyone thinks that so many people jump, Foxconn must be too dark, but we do not notice Foxconn employees about 370,000 people, according to the 12-hop suicide rate of less than four out of 10,000, and the national average suicide rate of fifteen out of 10,000, n even jump suicide rate is far lower than the national suicide rate, It can be seen that Foxconn 12-hop is actually a social problem, not just a business problem, we are too focused on the obvious fact but ignore the overall probability behind. Two days ago, 3 people were killed in the Boston bombing in the United States, all sorts of blessings on Weibo, but Afghanistan, Iraq and other countries are facing these problems almost every day, only because the media will not be all day to report the news there, and the daily attacks have paralyzed people's nerves, so we will only focus on the sharp Boston bombing, and indifferent to what happens every day in other parts of the country. In addition, some people around you to buy a lot of money, you may also want to join the stock market a try luck, and ignore the retail 8 to 1 flat 1 to earn the overall probability. You see a variety of entrepreneurial success stories, that they can also try to start a business, after all, the probability of success seems to be not low. But you don't know that those who are unsuccessful are less likely to be reported, but in fact they may be less than 1%.

So much so, in fact, is too bright accidental events will let us ignore the overall probability of the existence behind. When you see this data, don't be too emotional, you see data or events may only be a case, and can not represent the majority of the historical situation or the average situation, to find the silence of the user or data, avoid easy to make judgments and decisions. To rationally look at these accidental events, neither blindly follow, nor do not scoff, in the clear overall probability of the case, remove accidental factors, analysis of these accidental events behind the existence of some worthy of reference, so as to absorb their products or projects, So that their products or things to deal with May become the next "accidental event" in the market.

The tears of Looking back