How to Lie with Statistics
The book” How to lie with statistics” by Darrel Huff remains to be one of the best-known calls for critical thinking about statistics. The book distinguishes statistics and lying and how it tends to ignore the statistics construction process. For instance, usually, statistics are disseminated through research findings to attract media coverage. Previous studies have found that the availability of computers has facilitated the distribution of dubious statistics. The book is about the tricks used by news media to fool us about missing data. Therefore, this paper is meant to explore statistics language and the tricks that are employed to achieve sensationally, inflate and oversimplified statistics results.
From chapter one of the book, the most critical thing that Darrell talks about is how to use a sample with the built-in sample. He argues that if a sample too small, random or is not a representative of the whole data, then a study will be deemed less reliable or rather non-intelligent guess (Huff Darrel, 1954). Creating a good sample is crucial in statistics since it creates a basis for the conclusion of data. Huff argues that the only sample that gives reliable statistics results is a purely random one. To sum up, the point is for a sample to efficient it must possess two qualities and that is; it must be random and secondly, it must be large enough for statistical significance to be achieved.
Additionally, about sampling, Darrel states that one can employ stratified random sampling to avoid biased sampling. He elaborates that this is achieved by diving the population of study into subgroups according to their commonness (Darrel Huff, 1954). Thereafter a random sampling is done in within the subgroups. This type of sampling is described to be the most challenging since the strata are often subjective. Hence, for a survey to be effective, it must include three layers of sampling and they include; random population sample, questionnaires and random answers from randomly asked questions from the population. Therefore, it is important to note that purely random sampling often contributes to significant statistics results.
Another interesting thing that Darrel talks about is being unclear about the average that you use. The three types of averages mentioned are mean, median and mode. The arithmetic mean has to be evenly distributed among the total individual. The mode is an outcome that is the most viewed in a distribution outcome. Lastly, the median divides the distribution income into 50th percentile. Therefore, it is important to be keen about median, mean and mode during a statistical analysis to avoid obtaining skewed results.
Also, reliable studies always use statistically significant samples to obtain results. This is because small sample sizes give misleading results. According to Darrel, one should avoid reports, graphs and infographics that have not employed a sample to obtain their results. For instance, apart from numbers, words can also mislead the results or interpretation. This can be illustrated by the statement, 75% if the farms had electricity available (Darrel Huff, 1954). The statement is vague and does not explain in further details, for instance, it does not give the clear meaning of available. Therefore, statistical studies should employ statistical significant samples to avoid significant bias in their results. However, significant bias is essential for those producing new products and services.
Additionally, Darrel illustrates about margin errors in statistics. He says that since our answers normally lie between a certain range and therefore, it is not wise to make a big deal out of an insignificant result. According to Huff getting a 2% out of a sample of a 100 is not significant but instead, we should pretend that it is significant. This is because ranking is always open to marginal errors (Darrel Huff, 1954). No measurement can be perfectly accurate and therefore we have to take into account the standard error from the measurements of our data. Therefore, when you use smaller samples, the probability of obtaining a large variation is high. For instance, in an experiment of a 10-coin flip, there are fewer chances of getting 8 heads, but you are more likely to obtain 80 heads in 100 coin flips. Therefore, large samples should be used to avoid large standard errors.
Darrel Huff in his book also talks about how to lie in statistics using gee-whizz graphs and inaccurate pictures. Sometimes this deceptive method is referred to as semi-attached figures. This is the best way of convincing someone about something without proof, Darrel states that you only need to use figures to pretend that your results are reliable and based ion truths (Darrell Huff, 1954). Consequently, a semi-attached figure is plotted by picking two objects that seem similar but explicitly they are not and then draw a graph of comparison between them. Drawing a graph of comparison between two similar objects is the best way of convincing someone that your results are statistically significant.
Additionally, Huff demonstrates that one can convince someone about statistics results by using percentages to make comparisons. He clarifies that to achieve this one has to forget about what they are comparing with (Darrell Huff, 1954). For instance, use a shampoo that makes hair up to 60% shinier. From the results, it does not compare 60% of what, but eventually, the comparison is not specific to anything since we could compare it with anything. Therefore, the percentage is the most reliable method to convince someone about statistics figure since we do not need to compare the objects.
Another important aspect learnt from the book is about is correlation versus causation. Huff argues that confusing the two can lead to unjustified conclusions. Correlations are easier to found in most studies but they do not mean anything (Darrell Huff, 1954). This is because it can easily be fixed, for instance, adding decimal to answers; 49.9%, makes them appear more precise, certain and scientific. This is because an individual cannot be much less of 49 than 9. Huff illustrates post-hoc fallacy as an error that occurs due to simple misunderstandings. This is a tendency to assume a causal relationship between objects because they occur at the same time.
Additionally, the three types of correlations are demonstrated in further details, the first one is, correlation produced by chance, this a correlation resulting from different experiments using small samples. Secondly, covariation, it demonstrates a real relationship between two variables, however, their direction is unclear or impossible. Therefore, it is a right to conclude that correlations are necessary for a causality argument but sometimes they are insufficient on their own.
Another important aspect learnt from the book is the one-dimensional picture. This aspect involves using symbols in graphs. It is used to measure the growth of an object, for instance, a factory. Therefore, when the size of the factory image increases, it increases across all dimensions (Darrell Huff, 1954). A good example of it is a display of a difference in pay scale on a bar chart. Therefore, if you have one bar with one reading as 10 and the other 30, the ration 1:3 is a clear representation of the bar. Consequently, when the user sees these dimensions, they perceive an increase because the other dimension is forgotten. Therefore, it is important to use pictures in displaying dimensions of objects.
From the book, Darrel summarizes the basics of how to lie in statistics. He argues that these statistics lies do not always have an ill motive but they are caused unintentionally by simple mistakes. Therefore, according to Huff, most of the errors cause inflation and sensation to statistics meaning rather than to devalue and level them. Therefore, the author gives the right ways on how to defend oneself from bad statistics (Darrell and Huff, 1954). The following questions should be asked; who conducted this study and what was their motive behind it. This question is relevant because most researchers and companies tend to produce results that are favourable to them.
The second thing to do is be suspicious of the available data; both stated and unstated. A keen interest should be put on the type of sample that they used, whether they are accurate enough or can produce bias results (Darrel Huff, 1954). The third thing is to check on correlation-whether the samples are large enough to produce significant results. The study has to involve participants that were carefully selected through a purely random sample. The fourth thing to observe is if the standard errors included and the type of averages involved. All the mentioned elements must be accurate enough for the study to be effective.
Finally, according to Darrell Huff, one should check out on the missing data to identify whether the statistics are bad or good. The numbers should be thoroughly checked if the accurately lead to the obtained results. Additionally, false causality should also be observed.
In conclusion, from this book, I have been able to learn a lot about statistics. It is our right to stay alert and ask the right questions concerning statistical results to avoid being duped. Scientists, marketers and businesses will always adapt the scientific, accurate, honest and ethical way in presenting their information to the public. The book was worthwhile reading and makes sense to anyone interested in finding trucks about lying in statistics.
Works Cited
Huff, Darrel, and T. O. How. “How to Lie with Statistics. W W.” W. Norton & Co., Inc., New York (1954).