Decision making is an important business function which is prevalent within every process at every level of an organization. It is largely dependent upon support from accurate information and data to successfully maintain effective and rational decisions on the basis of analysis of data and information presented. The results of analysis become the foundation for the decision being made. However, the prevalence of bias and errors within the data can result in erroneous decisions due to the analysis providing skewed results. Data analysis is based on data and facts which are objective, but the analysis of data is sometimes biased and inaccurate due to heuristic errors (Gibbons, 2015). Heuristics are a short cut memory method or thumb rules that people use to make faster logical decisions, (Gibbons, 2015). This is a major risk associated with data driven decision-making rendering it inefficient until and unless adequate consideration for sensitivity of data to the buyers and its effect on the data is identified and rectifications are made accordingly.
Various forms of bias exist within every process and is easily identified within marketing and political campaigns, but may not always be so obvious. Certain types of bias may be difficult to identify and “multiple biases exist within a single data set” (Crawford, 2013). It is essential that every business leader educate themselves regarding prevalence and management of bias within data in depth rather than depending upon data scientists to provide information and guidance, as finally the “business leaders are responsible for the decisions they make on the basis of this data and need to face the repercussions for wrong decisions based upon biased data” (Crawford, 2013).
Confirmation Bias is a cognitive bias resulting from personal assumptions, opinions a hypothesis of individuals and unintentional or intentional desire existing to prove this to be true (Gibbons, 2015). The data becomes largely biased by confirmation bias existing during data collection or data being proved on the basis of what feels right rather than on the basis of true fact. It is essential that all statistical data collection and analysis be absolutely an objective process without the involvement of personal opinion emotions and providing minimal flexibility for data collection and processing. However, the inherent scope for wearing the process in terms of the variables used to calculate and the method of calculation itself can provide the scope for confirmation bias. It can result simply from collection of processing data by an individual who has been “processing it on a constant basis and is likely to have high levels of preconception with strong opinions and assumptions on the data processing and collection” (Morgan, 2019). For example, random sampling errors can be used in an extensive manner to misrepresent facts and figures or to control the output results acquired by manipulation of the input data set. Confirmation bias can also be transferred from business leaders for data analysts who are involved in the actual data processing expecting the reports to be in conformity with their opinions and expectations. Confirmation bias transfer can result in erroneous decision with extensive damage being “costly if not identified and controlled in the initial stages by ensuring that all individuals responsible for data processing are free from preconceived assumptions, opinions, and not under the influence of leaders having the capacity to enforce confirmation bias transfer”.
Selection bias results from input data being selected subjectively rather than in and objective man utilization of non-random data sets leading to statistical analysis results not to representation of the entire population. Selection bias can be the result of organizations failing to capture the entire picture to accurately represent the segment of organizations or a target group which is the deciding factor for the specific decision to be made. Data is critical for business decisions, and until data is accurate and with minimal contamination, the result can be inaccurate analysis resulting in inefficiency with even poor standards of quality being implemented. To minimize risk from selection bias, it needs to be ensured that input data selected is random with absolute representation of the entire population of the data.
Outliners are data averages with the result of the simple average greatly flawed due to presence of extreme data values measuring significantly above or below the normal range of values included or largely deviates from values within the normal distribution pattern. They can “dangerously alter the results with inexperienced data scientists or managers ignoring the impact such values can have on the average or mean figures apply the basic formula for mean calculation without analysis of input data out of habit or as a matter of procedure”. Outliers can be extreme enough to completely alter the results of an analysis, and if adequate consideration is not provided to minimize impact of search extreme data values by removing them before calculating averages or making a simple adjustment to bring them in alignment with the rest of the data distribution, the decision based on the output can result in serious damages. Large data sets may resist manual checking for outliers and may require specific measures being adopted to minimize impact from the existence of such a large data values within the data sets used for decision-making. Every organization need to adopt the most suitable procedure for handling outliers within the data set and introduce the process as it may not be always prudent to ignore outliers in certain organization such as insurance fraud. This is the reason behind various types of research, including medical findings reporting contrary results at different points in time and various promotional campaigns which are seemingly successful finally proved to be not very successful. “This is one of the most common biases in data analysis and results from use of descriptive statistics or data visualization leading to bias and wrong decision when working with big data” . For example, a promotional campaign may provide a minimal return on investment when incentivization of customer is based on erroneous conclusions. The greater and more in-depth analysis is essential for validating any trend which is visible for a marketing campaign to be sure that it is true by supporting in-depth analysis of all related factors. Ignoring this risk minimizing factor result in a promotional campaign reducing margins and in the long-term increasing the loss incurred from the field campaign instead of achieving the desired result of driving sales up and maximizing profit. Any initial trend means adequate validation for the supporting analysis through testing of control group measurements to ensure that the trend being visualized is correct. Current marketing analytic tools provide multivariate testing for adequate levels of aggregation been achieved through slicing and dicing of available data. This can however, result in averages which can be erroneous or misleading as it is not necessary to find a level of aggregation which does not need to be validated “if contradictions are not identified on an immediate basis”. In such cases the tendency becomes to follow your instinct to validate and use personal opinion to believe that the trend is true.
Therefore, it is extremely important that all data and analysis utilized for critical decision making within businesses is checked for accuracy and the entire process is controlled and conducted in a manner to minimize bias through implementation of processes and procedures to ensure elimination of bias in decision-making.