Measures are the building blocks that make up metrics and KPIs. They’re numeric and quantifiable, meaning they answer the “how much” and “how many” questions.
Dimensions are descriptive and qualitative, offering context to data. They help shape analysis and are often used to filter, group and slice data for aggregations and insights.
Nominal Scales
The type of scale you use to measure independent variables affects how the data is analyzed and what conclusions can be drawn from your research. Understanding the four primary scales of measurement – nominal, ordinal, interval and ratio – allows you to select the appropriate statistical analyses for your data.
Nominal scales categorize data into mutually exclusive groups without any inherent order or ranking, such as gender (male, female), eye color or product type. They also lack a numerical value for the categories, meaning that no arithmetic operations can be performed on these data, and only descriptive statistics such as the mode can be calculated.
On the other hand, ordinal scales allow for a rank order to be applied to data points, such as “highest,” “second highest” and “lowest.” Interval scales provide ordered categories with equal intervals but still lack a true zero point, so arithmetic operations such as addition, subtraction and multiplication cannot be performed on them. Finally, ratio scales combine properties from the other three types of scales and contain a true zero point, which enables you to perform all arithmetic operations on them.
Measures of Distance
The concept of distance is a general one that can be used to quantify many different things. It can be a physical length such as the distance travelled by an object, or it can be a measure of separation such as the Mahalanobis distance between two samples. It can even be an abstract notion such as the distance between feature vectors delivered by face or texture extraction algorithms.
A common measurement of distance is the Euclidean distance which is a scalar quantity that only has magnitude and not direction (as opposed to displacement, which does have direction). However, this method can be problematic when the data sets are high-dimensional because it suffers from the curse of dimensionality.
In such cases, other methods need to be used. For example, k-NN, a popular supervised learning algorithm, uses euclidean distance as its default parameter but it is important to understand that this may be a bad choice for your particular use case.
Measures of Time
Time measurement involves devising methods to count iterations of phenomena that repeat themselves at regular intervals, such as the rotation of the Earth on its axis and shifts in the phases of the moon. It can also be used to quantify the duration of events or the gaps between them. This is called chronometry.
Although physicists have never managed to define time itself, they have developed extraordinarily accurate ways of measuring it. They use clocks and calendars to record an instant or a date, and they use time interval and frequency (the rate at which a repeated phenomenon occurs) to determine durations of events or the gaps between them.
The standard unit of time is the second, which is one of the seven base units of the International System of Units. It is based on the vibrations of caesium atoms, making it an extremely precise and reliable standard. It is used worldwide to synchronize clocks.
Measures of Emotion
Emotion is a neurobiological process, a phenomenological experience and feeling, and a perceptual-cognitive process. Researchers seek to measure all of these dimensions of emotion in their research, and they use different techniques to do so. These third-person techniques include behavioral, neural, and physiological measurement methods that are objective in nature. Alternatively, researchers can utilize first-person self-report methodologies to assess subjective experience and feelings.
While previous studies have found significant associations between language measures of valence and discrete emotions and self-reports, these analyses have been constrained by their scope. This study expands the scope of those investigations and tests for associations between language operationalizations and other measures that are often used to assess emotion in three large, multimodal datasets.
Using a new difference scaling method, we find that the MoVEE correlates moderately with quality-intensity emotion rating scales, including the PANAS. However, correlations with facial cues are less robust. This may be due to norms against facial displays of negative feelings, which reduces the likelihood that these signals will seep into language.