A very interesting, and broad (compared to our more statistics-specific discussions in MBAG 8541A) topic is touched in this book: http://time.com/4477557/big-data-biases/. The basic point is: data analytics (from basic descriptive statistics, to inferential statistics, to econometrics and bid data analysis) is subject to all the normal human biases the analysts might possess. The problem, however, is that big data now leads the charge toward behaviour and choice automation.
The book review focuses on ethical dimensions of this problem. There is also a remedial cost dimension - with automated behaviour based on biased algorithms, corrective action cannot take place ex ante automated choice, but only either ex ante analysis (via restricting algorithms) or ex post the algorithm-enabled action takes place. Which, of course, magnifies the costs associated with controlling for biases.
One way or the other - the concept of biased algorithmic models certainly presents some food for thought!
Dwenger, Nadja and Lohse, Tim paper “Do Individuals Put Effort into Lying? Evidence from a Compliance Experiment” (March 10, 2016, CESifo Working Paper Series No. 5805: http://ssrn.com/abstract=2764121) looks at “…whether individuals in a face-to-face situation can successfully exert some lying effort to delude others.”
The authors use a laboratory experiment in which “participants were asked to assess videotaped statements as being rather truthful or untruthful. The statements are face-to-face tax declarations. The video clips feature each subject twice making the same declaration. But one time the subject is reporting truthfully, the other time willingly untruthfully. This allows us to investigate within-subject differences in trustworthiness.”
What the authors found is rather interesting: “a subject is perceived as more trustworthy if she deceives than if she reports truthfully. It is particularly individuals with dishonest appearance who manage to increase their perceived trustworthiness by up to 15 percent. This is evidence of individuals successfully exerting lying effort.”
So you are more likely to buy a lemon from a lemon-selling dealer, than a real thing from an honest one... doh...
Some more ‘beef’ from the study:
“To deceive or not to deceive is a question that arises in basically all spheres of life. Sometimes the stakes involved are small and coming up with a lie is hardly worth it. But sometimes putting effort into lying might be rewarding, provided the deception is not detected.”
However, “whether or not a lie is detected is a matter of how trustworthy the individual is perceived to be. When interacting face-to-face two aspects determine the perceived trustworthiness:
- First, an individual’s general appearance, and
- Second, the level of some kind of effort the individual may choose when trying to make the lie appear truthful.
The authors ask a non-trivial question: “do we really perceive individuals who tell the truth as more trustworthy than individuals who deceive?”
“Despite its importance for social life, the literature has remained surprisingly silent on the issue of lying effort. This paper is the first to shed light on this issue.”
The study actually uses two types of data from two types of experiments: “An experiment with room for deception which was framed as a tax compliance experiment and a deception-assessment experiment. In the compliance experiment subjects had to declare income in face-to-face situations vis-a-vis an officer, comparable to the situation at customs. They could report honestly or try to evade taxes by deceiving. Some subjects received an audit and the audit probabilities were influenced by the tax officer, based on his impression of the subject. The compliance interviews were videotaped and some of these video clips were the basis for our deception-assessment experiment: For each subject we selected two videos both showing the same low income declaration, but once when telling the truth and once when lying. A different set of participants was asked to watch the video clips and assess whether the recorded subject was truthfully reporting her income or whether she was lying. These assessments were incentivised. Based on more than 18,000 assessments we are able to generate a trustworthiness score for each video clip (number of times the video is rated "rather truthful" divided by the total number of assessments). As each individual is assessed in two different video clips, we can exploit within-subject differences in trustworthiness. …Any difference in trust-worthiness scores between situations of honesty and dishonesty can thus be traced back to the effort exerted by an individual when lying. In addition, we also investigate whether subjects appear less trustworthy if they were audited and had been caught lying shortly before. …the individuals who had to assess the trustworthiness of a tax declarer did not receive any information on previous audits.
The main results are as follows:
- “Subjects appear as more trustworthy in compliance interviews in which they underreport than in compliance interviews in which they report truthfully. When categorizing individuals in subjects with a genuine dishonest or honest appearance, it becomes obvious that it is mainly individuals of the former category who appear more trustworthy when deceiving.”
- “These individuals with a dishonest appearance are able to increase their perceived trustworthiness by up to 15 percent. This finding is in line with the hypothesis that players with a comparably dishonest appearance, when lying, expend effort to appear truthful.”
- “We also find that an individual’s trustworthiness is affected by previous audit experiences. Individuals who were caught cheating in the previous period, appear significantly less trustworthy, compared to individuals who were either not audited or who reported truthfully. This effect is exacerbated for individuals with a dishonest appearance if the individual is again underreporting but is lessened if the individual is reporting truthfully.”