Business Management Dynamics

Double Blind Peer Reviewed - Open Access Jounal

Home  |   Contact Us

 

ISSN: 2047-7031

bmd Business and Management Dynamics bmd
ISSN: 2047-7031  
Volume  8   Issue 6  2018  
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Article Abstract
Testing the Small Size Effect Bias for Benford Screening: The False Negative Signaling Error
Keywords:  T. Hill, Non-Conforming Benford Datasets, False Positive Screening Error
Frank Heilig & Edward J. Lusk
Recent research has provided important information on the effect of partitioning large datasets that are likely to be Conforming to the Newcomb & Benford profile. This research documents that at some point, as the sample size is systematically reduced, sub-samples randomly drawn from Conforming datasets, test to be Non-Conforming. This has been termed a False Positive Screening Error [FPSE]-Incorrectly classifying a Conforming dataset as Non-Conforming; otherwise said: Failing to detect the True State of Nature of the data generating process and so in, the audit context, to incorrectly make the decision to investigate the dataset as one that may have been manipulated to a nefarious end. These research reports beg the question that motives our research-to wit: Is there such a partitioning effect if the dataset is Non-Conforming in nature? This is termed a False Negative Screening Error [FNSE]: Accepting as Conforming a Non-Conforming dataset and so failing to effect an Extended Procedures examination in the audit context when one would have been prudent. Method For control purposes, we used: (i) the same Decision Support System as was used in our previous research to screen the various sampled partitions, and (ii) the same partitioning algorithm to create the randomly drawn sub-samples. Results We find no evidence that the FNSE is produced at a rate that would usually be considered as counterproductive to the effective and efficient execution of the audit. Further, we used a simple Bayes filter to identify those Non-Conforming datasets that are in the definitive end of the Non-Conforming scale. In this case, there is still a FNSE jeopardy, albeit somewhat reduced. Impact These results add to the information on dataset partitioning or accrued from the onset of the data generating process. We document that there is a jeopardy difference between the FPSE and the FNSE. While Conforming datasets tend to be affected by sampling and incorrectly signal investigations at reduced sample sizes (FPSE), Non-Conforming datasets do not show the same tendency-i.e., to incorrectly decide not to investigate (FNSE).
23-34   |  View PDF