Actually, there is no problem per se — but there can be. Statistics are infamous for their ability and potential to exist as misleading and bad data.
Misleading statistics are simply the misusage - purposeful or not - of a numerical data. The results provide a misleading information to the receiver, who then believes something wrong if he or she does not notice the error or the does not have the full data picture.
As an exercise in due diligence, we will review some of the most common forms of misuse of statistics, and various alarming and sadly, common misleading statistics examples from public life. Statistical reliability is crucial in order to ensure the precision and validity of the analysis.
To make sure the reliability is high, there are various techniques to perform — first of them being the control tests, that should have similar results when reproducing an experiment in similar conditions. However, the telling of half-truths through study is not only limited to mathematical amateurs.
A investigative survey by Dr. Daniele Fanelli from The University of Edinburgh found that There are different ways how statistics can be misleading that we will detail later. The most common one is of course correlation versus causation, that always leaves out another or two or three factor that are the actual causation of the problem. Did we forget to mention the amount of sugar put in the tea, or the fact that baldness and old age are related — just like cardiovascular disease risks and old age?
So, can statistics be manipulated? They sure can. Do numbers lie? You can be the judge. Remember, misuse of statistics can be accidental or purposeful. While a malicious intent to blur lines with misleading statistics will surely magnify bias, intent is not necessary to create misunderstandings.
The misuse of statistics is a much broader problem that now permeates through multiple industries and fields of study. Here are a few potential mishaps that commonly lead to misuse:.
The manner in which questions are phrased can have a huge impact on the way an audience answers them. Specific wording patterns have a persuasive effect and induce respondents to answer in a predictable manner. These two questions are likely to provoke far different responses, even though they deal with the same topic of government assistance. The latter two examples of the original questions eliminate any inference or suggestion from the poller, and thus, are significantly more impartial.
Another unfair method of polling is to ask a question, but precede it with a conditional statement or a statement of fact. A good rule of thumb is to always take polling with a grain of salt, and to try to review the questions that were actually presented. They provide great insight, often more so than the answers.
The problem with correlations is this: if you measure enough variables, eventually it will appear that some of them correlate. As one out of twenty will inevitably be deemed significant without any direct correlation, studies can be manipulated with enough data to prove a correlation that does not exist or that is not significant enough to prove causation. Any sensible person would easily identify the fact that car accidents do not cause bear attacks.
Each is likely a result of a third factor, that being: an increased population, due to high tourism season in the month of June. It would be preposterous to say that they cause each other It is easy to see a correlation. But, what about causation? What if the measured variables were different? Clearly there is a correlation between the two, but is there causation? Many would falsely assume, yes, solely based on the strength of the correlation.
Tread carefully, for either knowingly or ignorantly, correlation hunting will continue to exist within statistical studies. Subjects were shown a film depicting multiple car accidents. Using language to influence survey answers and results is just one example of selection bias. Selective bias often occurs when chosen samples or data are incomplete or cherrypicked to influence the perception of - and even skew - statistics and data. Smaller sample sizes almost guarantee alarmingly significant results.
Always beware of extreme results, and never accept percentages at face value. In the words of biochemistry researcher Ana-maria Sundic :.
Correlation and causation warrant plenty of suspicion because researchers - and consumers of said research - fall prey to:. Tyler Vegihn compiled some funny misleading statistics examples to prove this exact point:. This graph depicts a compelling correlation between the number of people who drowned falling into a pool and the number of movies Nicolas Cage appeared in:. Another shows a correlation between the number of people who died by becoming entangled in bedsheets, with cheese consumption:.
Probably not. Data visualizations turn raw numbers into visual representations of key relationships, trends, and patterns. One popular example from the news is the Terri Schiavo case , a right-to-die legal case in the U. A glance at this graph suggests that when compared to Republicans and Independents, 3 times more Democrats agreed with the court.
The truncated graph and tampered Y-axis starting at 50 instead of 0 distort the data, and lead you to believe an exaggerated idea about a certain group. The intervals and scales. Check for uneven increments and odd measurements use of numbers instead of percentages etc. The complete context and other comparative graphs to see how similar data is measured and represented. They create shocking headlines that attract swarms of traffic but provide flawed insights at best.
Instead of helping you navigate through detours, potholes and pitfalls, they knowingly- or unknowingly - steer you right into them. Research is expensive and time-consuming. Check who is sponsoring it, weigh their bias on the topic and how they might benefit from results.
Are they a B2C company with a product? A consulting service? Collecting data from too small a group can skew your survey and test results. Small samples underrepresent your target audience.
They can lead to misleading statistics that give you a faulty idea of customer satisfaction and product preferences. Sample size is especially important if you analyze results in terms of percentages instead of exact numbers. If you tested 10, people, that percentage is a pretty convincing reason to develop that version.
But if you tested only 20 people, that means only 12 are interested in the idea. Twelve is too small a group to show you that the new feature will be worth your investment. Small group sizes can also lead to biased sampling. The smaller the group, the less likely it is to represent the diverse demographics of your customer base. An ideal sample size depends on many factors, like your company and the goals for your project. Using a third-party tool helps you reliably assess your sample size without having to figure out the calculations on your own.
Users enter their expected conversion rate and the percent change they are looking for. The way you word survey questions can also be a source of misleading statistics. A recent UK study shows that the way you phrase a question directly affects how a person answers. One example is survey questions that ask for confirmation. Essentially, you are including the answer in the question.
Check your surveys for manipulative wording that might lead respondents to give a particular answer. A few examples of influential phrasing include:. Check for leading language by asking co-workers to review surveys before sending to customers. Ask what parts of your questions, if any, suggest how they should respond.
Confirmation bias is when you have a set result you want or expect to see, so you look at only data that affirms your belief. Companies are most susceptible to this phenomenon when a single employee is giving a presentation. Whether the employee realizes it or not, they may not be providing a full picture of the data due to their own views—and that can lead to poorly informed decisions. To support her claim, she shows that very few customer support calls mention this feature.
As it turns out, she was looking at calls from only the last six months. When analyzing support calls from long-term customers, the product team sees a much higher percentage bringing up issues with the Favorites feature. Everyone has unconscious biases. But not everyone has the same ones. If an employee comes to you with a proposal, have another team member review the project idea and the presentation. Each person will approach the data differently, so someone will likely realize whether the data is skewed toward one perspective.
Offer training to help employees become aware of their biases. This is especially important when it comes to internal hiring and employee development decisions.
0コメント