What is Ordinal Data?
Ordinal data is defined as a qualitative scale of measurement where there is a meaningful order between the points on the scale, but the difference in intervals between the points is not measurable in numbers.
For example, a customer service survey question asks respondents to measure their service between ‘very good’, ‘good’, ‘neutral’, ‘bad’, ‘very bad’. Here, you can see that there is a clear order going from ‘very good’ to ‘very bad’. There are also meaningful differences between the options, but the gap between each is not measurable in numbers. Therefore, it is also qualitative in nature.
To recap and compare, data on scales of measurement are nominal, ordinal, interval and ratio. Here, nominal data has no order or intervals and only plain self-sustaining labels. Ordinal data on the other hand does have order but not measurable intervals. Then interval data has order plus measurable intervals but no true zero point. And ratio data has all the previous plus a true zero point.
Ordinal data is commonly used in rating scales on feedback forms. This data can then be analyzed to understand trends in feedback and where action is needed.
For example, borrowing from the previous example, ordinal data from 100 customer feedback surveys may show that 45/100 responses show “good”, followed by 30/100 that say “very good”, 10/100 say “bad” and 15 say “very bad”. Based on this data trend, it can be concluded that 75% of the customers have a positive view of the customer service. But 25% of bad feedback is still a significant number for any business and further enquiry may be needed to gain better understanding of their pain points and rectify them.
Key Characteristics of Ordinal Data
Let us now dive into getting a clearer understanding of the key characteristics of ordinal data:
- Categorical nature with an order
Unlike nominal data where categorical representation of labels are self-containing with no order among them , ordinal data has categories with a clear order. For example, when asked about an audience’s intake of food with vitamin C on a daily basis, the ordinal data points may be represented in the order of “high”, “medium”, “low”.
- Unquantifiable interval spaces
The intervals between the categorical points in ordinal data are not mathematical/ quantifiable, and are instead qualitative.
For example, in a pleasure scale while having ice cream, ranging from level 1 (very low pleasure) to level 5 (very high pleasure), we know that a level 4 indicates more pleasure than a level 3, but the actual sensation gap between any of the points is not quantifiable mathematically.
- Numeric and non-numeric representation
Ordinal data as such can be represented on a scale using labels, which are non-numeric, and numbers. However, the numbers are merely label-replacements for the sake of representation and don’t represent a true value.
For example, academic performance can be categorized as “Fail,” “Pass,” “Merit,” and “Distinction,” or alternatively, these can be numbered as 1, 2, 3, and 4 on a scale. That is all, their meaning remains the same, which is qualitative.
Therefore, for ordinal data, numbers may simply be used for better understanding or representation.
Ordinal Data Examples and Sources
Ordinal data can be collected from various sources across different fields and contexts. Here are some common sources:
- Market research surveys:
Market research surveys typically used ordinal data to collect opinions from a select group of respondents based on their match for certain criterias that qualify them.
A simple example can be that of a product research where they are asked questions on how likely they are to use a certain product feature on an ordinal scale such as “very less likely”, “less likely”, “moderate”, “likely”, “very likely”.
- Customer/ user feedback collection:
Customer and user feedback is another space where ordinal data is often used to measure their level of satisfaction with a product, service or quality of interaction with the brand be it online or offline.
For example, a car brand may collect online feedback on customer experience after their visit to their showroom, and the ordinal scale here may contain points such as “very good”, “somewhat good”, “somewhat bad”, “very bad”. Typically the questions then branch off to other rating questions on specific parts of the experiences or simply open ended questions to gather their overall opinion.
- Educational Assessments:
Education systems use ordinal data often to rate qualitative performances where exact marks are not possible to capture.
For example, teachers may rate students on their level of attentiveness during classes on a scale of “very attentive”, “quite attentive”, “quite distracted”, “very distracted”.
- Healthcare systems
Hospitals and clinics often see patients being asked to respond using ordinal data as most such responses are qualitative.
For example, a mental health professional may ask a patient to respond on how they are feeling to understand their stress levels on a scale of “very stressed”, “somewhat stressed”, “somewhat relaxed”, “very relaxed”.
- Economic and government classification
In economics, ordinal data is a common occurrence where people may be classified based on income such as “high income group”, “medium income group”, “low income group”, “below poverty line”.
In government and civics, quality of life is often a space where ordinal data may be used due to its level of subjectivity. For example, a civic survey attempting to understand the quality of life and happiness in an area may use a scale such as “very happy”, “happy”, “neutral”, “unhappy”, “very unhappy”.
Best Practices for Collecting and Managing Ordinal Data
Collecting and managing ordinal data requires careful consideration to ensure data quality, accuracy, and meaningful analysis.
Here are the key best practices for collecting and managing ordinal data in 2024:
- Design clear and unambiguous data points in surveys
It is always important to ensure that each category data point is clear and unambiguous, avoiding overlap in meaning. This helps respondents understand the choices and reduces confusion.
Arrange the categories in a logical and meaningful sequence such that the order becomes obvious.
For example, in a satisfaction survey, order responses from “Very Dissatisfied” to “Very Satisfied.”
- Standardize and test
Use standardized scales across surveys or studies to enable comparability over time and across different groups. For example, using a consistent 5-point scale for measuring agreement level, with the same response options.
Furthermore, it is always good to conduct a pilot test of the survey or questionnaire to identify potential issues around interpretation or understanding of the ordinal scale response options.
- Ethics and Respondent Comfort
Like with any data collection effort, ensure that respondents know their participation is voluntary and that they can opt out at any time, especially when collecting sensitive information (e.g., health status, socioeconomic status).
Also provide them a clear understanding of how the collected data will be used and explain how you ensure respondent confidentiality and anonymity.
- Cultural considerations
Be very aware of culturally sensitive questions that may affect the interpretation of ordinal scales or even the questions themselves. Adapt questions and response scales to be culturally appropriate and sensitive.
Depending on the geography, it is always a good idea to hire/ consult a specialist locally to help you frame questionnaires or at least to verify them.
- Data management and analysis
Implement checks to ensure data integrity, such as preventing duplicate responses and addressing missing data. Use secure data storage solutions to protect data from unauthorized access.
Use suitable statistical techniques for analyzing ordinal data. Non-parametric tests, such as the Mann-Whitney U test or Spearman’s rank correlation, are often appropriate due to the lack of consistent intervals between categories.
When reporting results, use appropriate visualizations, such as bar charts or pie charts, to display ordinal data. Avoid using misleading representations that imply equal intervals between categories.
