Decision Making as Output and Bounded Rationality

  The classical economics theories proceed on the assumption of rational agents. Rationality implies the economic agents undertake actions or exercise choices based on the cost-benefit analysis they undertake. The assumption further posits that there exists no information asymmetry and thus the agent is aware of all the costs and benefits associated with the choice he or she has exercised. The behavioral school contested the decision stating the decisions in practice are often irrational. Implied there is a continuous departure from rationality. Rationality in the views of the behavioral school is more an exception to the norm rather a rule. The past posts have discussed the limitations of this view by the behavioral school. Economics has often posited rationality in the context in which the choices are exercised rather than theoretical abstract view of rational action. Rational action in theory seems to be grounded in zero restraint situation yet in practice, there are numerous restra

Tests of Statistical Similarity, Behaviour and Aftereffects


Quantification and objectification of compatibility is perhaps a defining feature in navigating the Industry 4.0.  In choice of vendors, franchisees, licensees, franchisors, licensors, customers etc. a key to long last partnership lies often in the compatibility of characteristics between them. Customer and producer journeys in multi sided platform markets evolve out of compatibility development among the various partners. Therefore the platform objective would to sync these mutually similar compatible attributes between the players on the platform.

Often a solution to swifter payback period entails following a time tested model of franchising or licensing. In identifying the right vendor, client or partner, a firm often would turn a basic meeting ground of quantitative parameters like financial credibility of the target partner, experience in the relevant field etc. with suitable weightages. In addition, there is a judgmental call based on perhps on gut or qualitative feel whether the target is right candidate or not. Depending on the market structure geared towards the buyers or supplier side, the other party too would seemingly examine quantitatively or judgmentally the presence of similarity in attributes towards making a successful journey.  The applications range from franchising, distributor networks, logistics agency, licensing, export agency among others.

Nonetheless, it is incontestable, that we live in a universe progressively centred by computable thinking. The human enterprise is to objectify each and every attribute human or otherwise. Irrespective of the merits or otherwise, this unidirectional movement is conceivably acknowledged. Therefore in building the partner base for any firm, quantitative approaches need to be structured on the vendor side. Statistical tools are freely to be used. It is not uncommon to find a few prophets of statistical infallibility argue for increasing sophisticating of statistical tools to get better results. It is moot to determine a linkage between statistical sophistication and increased reliability, validity and credibility of statistical tests. It would not be unreasonable to conclude yielding of diminish returns beyond certain level of input output analysis something which this piece highlights.  

In the instances discussed, Dice coefficient is a popular statistical measure that is being used. Serenson-Dice coefficient as it is known among other names is a tool for statistical measure of similarity. A university has developed a new product for which it has obtained a patent. To commercialize it can either choose to use in-house or licensee it to a client. In the process of identifying the right target client, Dice coefficient might yield greater utility in testing the similarities between the licensor and the target licensee. Indubitably, the challenge first is quantify the attributed for which the similarity has to be tested. Further given the results, the interpretation of the same is human. It is perhaps to use in conjunction with other parameters like citation analysis etc.

The applications of statistical similarity tests like Dice coefficient, Jacquard index etc. go beyond the ones highlighted above. Apart from use in vendor and customer selection they have increasing applications in what is being described as Industry 4.0.  Many critics posit that ancient texts are not composed by a single author but composed by multiple authors at different points of time. A good way would be to test writing styles and the vocabulary being used through the text. If there is similarity of vocabulary through the text, it might be of single author else might yield to speculation of being an agglomeration several parts being added at different points of time. In image processing and identification, Dice coefficient seems to have lot of applications. Going forward, vacation and tourism firms might begin using data from customers, hotels, airlines, cruises, tourism spots etc. run a similarity algorithms using Dice coefficient and its developed variants, identify a right tourism spot for you. If someone wants to purchase a car or a bike, there might be series of similarity tests run to find the right choice. A more unnerving would be a marriage market wherein compatibility tests would be performed using several attributes to decide whether the marriage will last or not. This would be quite a journey from astrology to statistics or astrological numerology to statistical numerology.

Yet these bring about numerous challenges. Statistical tools are merely tools. They give results based on the data. The first challenge is to build the quantitative attributes. This entails a journey of quantifying the qualitative. Several attributed which be intrinsically difficult to quantify might have to undergo a conversion process. With advancements in imaging technologies etc. even human images and at different points of time and different contexts can be assigned a numerical meaning. This of course would be hoary and might not be something palatable to human interest and desire. Technology need not be advanced for these analysis. Even images from mobile phones or CCTVs could be of good use. Errors in defining and quantifying attributes might pose new risks creating perhaps new class of risks. More than errors, the biases that creep in in designing algorithms through machine learning, deep learning, artificial intelligence might lead to unintended consequences. Construction of variables and quantifying the same have been a function of human endeavour. This is increasingly transferred to machines. So is the interpretation. The results while perhaps funny or hilarious in some circumstances, might lead to horrifying consequences.

An instance would suffice. Online ads are tailored to customers based on the past web browsing behaviour. The data from the past browsing behaviour is fed into systems and maybe let us say, Dice coefficient or similarity tools are used to judge the preferences of particular web user and therefore customised ads are tailored to his or her needs. Recently, a web user complained to railways that there appeared a lot of sexual or porn ads whenever he logged in to IRCTC to book his train tickets. He added, such ads were embarrassing. The reply of course was classic. The platform is built on delivering ads to customers based on their past browsing preferences and thus higher likelihood for clicking such ads. In the event of failure to delete cookies, these ads would be inevitable. Implied was the complainant perhaps used to browse lot of porn and therefore his failure to delete the cookies meant such porn ads keep appearing on the sites he browses given the higher likelihood of clicking the same. It was of course funny and downright embarrassing for the complainant. In his over enthusiasm to put IRCTC on the mat, his private behaviour was made public. It might just be an embarrassment but going forward there can be instances wherein these can lead to horrifying results.

While Dice Coefficient and other measures of statistical similarities would gain increased distinction in human pursuit to quantify and objectify every single trait human, animate, inanimate or otherwise, assembly and elucidation of the same might lead to numerous unintended externalities.

Comments

Popular posts from this blog

Decision Making as Output and Bounded Rationality

The Chicken-Egg Conundrum of Economics

A Note on Supply-Demand Dynamics