Issue 14608: Add support for Scale of Measurement
Issue 15011: operation attribit of direct measure class
Issue 15012: OCL expr. of BinaryMeasurement
Issue 15014: Inconsistency between specification and provided Meta Model
Issue 15015: Modeling names of model elements
Issue 15016: Threshold values for dimensional measures
Issue 15017: Measure audits
Issue 14608: Add support for Scale of Measurement (smm-ftf)
Click here for this issue's archive.
Source: CAST Software (Dr. Bill Curtis, b.curtis(at)castsoftware.com)
Nature: Uncategorized Issue
Severity:
Summary:
Summary: Scales of Measurement are a fundamental property of a measure describing the allowable mathematical transformations that can be legitimately applied to it. Discussion: Attached are some URLs that will give you an overview of Scales of Measurement. They are a fundamental property of a measure describing the allowable mathematical transformations that can be legitimately applied to it. Since they therefore determine the statistical methods that can be applied to the measure, they need to be treated as a first level attribute of a measure. http://www.math.sfu.ca/~cschwarz/Stat-301/Handouts/node5.html http://en.wikipedia.org/wiki/Scale_of_measurement (not always written in easily understood terms) http://www.wadsworth.com/psychology_d/templates/student_resources/workshops/stat_workshp/scale/scale_01.html (tutorial - behavioral science view) Summary of change: · Add enumeration MeasurementScale o Add enu ordinal, nominal, interval and ratio. · Add attribute scale:MeasurementScale to Measure class o This defines the scale of Measurement for measurements produced by the measure.
Attached are some URLs that will give you an overview of Scales of Measurement. They are a fundamental property of a measure describing the allowable mathematical transformations that can be legitimately applied to it. Since they therefore determine the statistical methods that can be applied to the measure, they need to be treated as a first level attribute of a measure. http://www.math.sfu.ca/~cschwarz/Stat-301/Handouts/node5.html http://en.wikipedia.org/wiki/Scale_of_measurement (not always written in easily understood terms) http://www.wadsworth.com/psychology_d/templates/student_resources/workshops/stat_workshp/scale/scale_01.html (tutorial - behavioral science view) Summary of change: • Add enumeration MeasurementScale o Add enum ordinal, nominal, interval and ratio. • Add attribute scale:MeasurementScale to Measure class o This defines the scale of Measurement for measurements produced by the measure. Discussion: From Larry Hines: With respect to 14608: First, scale is mostly redundant information. As such it could be an operation defined for Measure, but it should be an attribute. This operation would always return ‘ratio’ for RatioMeasures. It would always return ‘interval’ for DimensionalMeasures. The tricky part is to separate ‘nominal’ and ‘ordinal’ for Ranking measures. We could have an optional attribute for that. It wouldn’t hurt, but it is not needed to convey the meaning of a measurement. From Alain Picard: After reviewing Larry’s comment on his vote, I have to agree that it will be better to defer this issue to the RTF and to better implement a real solution to the issue. Disposition: Deferred
The operation attribute is declared to be of type Operation while this type is not specified in the document. Moreover, e.g. in Figure 5, it is declared to be of type string.
The OCL expression related to the BinaryMeasurement class refers to the context RatioMeasurement instead of BinaryMeasurement
Comparing the specification and the provided meta model, numerous inconsistencies between both appear. For example: • Naming differentials: Most of the meta classes with an underscore in their name mentioned in the specification document are named differently without the underscore. • Some association ends are missing in the provided model, e.g. categoryElement for the association between SMMCategory and Measure. (In our opinion, this association should be called measureElement to separate from the reflexive Association of the SMMCategory class having the same name.) • Numerous multiplicity values in the meta model differ from the specification. For example, the association between the SMMModell class and the SMMElement class has the upper bound value 1. • The subclasses of CollectiveMeasure (AdditiveMeasure and MaximumMeasure) mentioned in the specification do not occur in the meta model.
The chosen way of modeling names for model elements is not done straightforward and hampers the practical usage of the meta model. The SMM meta model contains several classes having an attribute called name. Moreover, all meta classes are derived from the abstract meta class SMM_Element, which should possess such an attribute according to the specification. Hence, this attribute can be considered of being modeled redundantly. Regarding the provided EMOF model, the SMM_Element does not have any attribute so that the specified attributes name, short_description and description are missing. According to best practices, we recommend to introduce a new class called SMM_NamedElement as a subclass of SMM_Element and as the base class of all meta classes having a name attribute. This facilitates to determine the name of a model element.
It is not stated clear how to assign threshold values to dimensional measures. On page 9 of the specification document, it is mentioned that a measure has a range which can be interpreted as “the set of possible measurement results”. However, there is no corresponding attribute mentioned in the specification or meta model. We suggest to add two attributes lowerThreshold and upperThreshold to the DimensionalMeasure meta class to model a measure’s range.
In some cases, it is helpful to compose several measures to a single audit to run these measures at once without the necessity to select this measures manually prior to each computation. Although it would be possible to model such as relationship using a subclass of the MeasureRelationship class, we recommend to introduce a class Audit that simply has a association to the Measure class to model the membership of a measure to a specific audit.