The CogentQC systems started with a sampling methodology, and then gradually evolved into the full-scale QC products that we market today. The five kinds of samples discussed below are the bread and butter of the Cogent system's statistical sophistication. They allow our users to establish data sets that both specific and mathematically useful. These are only brief explanations of each type of sample. More comprehensive information is provided to our clients on the system's help screens.

Statistical Sample 

The Statistical Sample is a random sample designed to estimate Overall Loan Quality. The Statistical Sample is the most basic level of sampling possible. It draws on a random basis from the entire loan population. To read more about Cogent’s statistical sampling methodology, click here.

Briefly, when asked to draw a statistical sample, the system computes suggested Statistical Sample size based on achieving 2% Statistical Precision (at a one-sided 95% Confidence Level). This computation is based on: (1) the Number of Loans Available For Sampling; and (2) the most recent Quarterly Defect Rate. By taking these variables into account, the system is able to determine how many loans must be sampled to guarantee precise results. For more information on statistical terminology such as “precision,” and “confidence level,” click here.

Stratified Sample 

The Stratified Sample is a discretionary random sample designed to address Origination Source Risk (the loan quality risk associated with source category). Loans from one Origination Source may be inherently more risky than loans from other Sources, and so you may want to oversample that Source. For example, wholesale lending typically involves more defects than retail lending, and it may be worthwhile to determine whether that is the case for your organization.

The stratified sample is the second layer of sampling. It is designed to facilitate comparisons between origination sources. It can correct for biases that emerge in simple statistical sampling by weighting particular sources appropriately, thus compensating for their inherent risk differentials.

Source Unit Sample 

Source Unit Samples provide randomized selection using selection criteria which are more fine-grained than stratification provides. In this Sample Group, the population of loans is extracted from the intersection of an Origination Source (such as Retail, Wholesale, Correspondent) with an Origination Unit (such as a Branch, Region). Thus, a Retail Branch Source Unit Sample is one selected from a population of loans sorted by all Retail Branches in the population. This type of sample is most useful when you want to examine loan quality of specific operational units (such as a Branch or group of Branches).

Targeted Sample

Targeted Samples comprise the most flexible Sample Group you can generate, since the User defines the exact population from which the sample will be selected. The process (of defining the population) involves constructing a Query, essentially a set of rules you want the system to use in defining your population. Queries can range from quite simple to very complicated, but Users have almost unlimited flexibility in defining the population. Advanced users are fans of targeted samples as they allow QC departments to leverage the software and find the most specific results.

Manual Sample

Manual Samples are a special category of Targeted Samples. The name is relatively self-explanatory; with this sampling option, the user directly selects the loans that he or she wants to sample.