How to choose a model to organise your data
What is data modelling?
Data modelling is about how you structure your database. A simple example would be deciding which columns you would use in a spreadsheet, the different sheets and how you link between them. Below we discuss data modelling in more depth, but a word of warning - data modelling is a complicated field and if you have no prior experience it’s a good idea to talk to an expert before embarking on data modelling yourself.
Why model your data?
Your model is the structure through which you can ask what relations there are between different aspects of the data. It requires choice over how you categorise the data - into what buckets you decide to put data with certain characteristics, and how you allow those categories to relate to each other. What you can find out about your data set depends on how you group together certain elements of the data. Making the right choice at the beginning will save you a lot of work in the long run.
How to do it
With your model you want to create a structure that reflects the real life system that the data is about. Organising your data goes hand in hand with data collection. If the “rules” of the model don’t accommodate the data you’re collecting, it can require you to “break” your model, or to start over. Say you’re studying a corporate structure, and you create a data model that lists an individual’s name against their job title. If your data model is set up to reference only one job for a particular person, when you come to Anthony Hu who is the Events Coordinator and Accounts Assistant, you won't be able to capture this without "breaking" your data model - you might create duplicate rows which would lead to duplicate Anthonys. Building on this example, if you wanted to also consider the dates between which someone held a job, you should structure your data model to make this data accessible. In that case you could add a separate sheet to hold the job's data and the start and end dates. You would also want to add an identifier for each job that links it to the person in your other sheet.
Choosing the right type of model for you
This process is the basis of most data modeling, and specifically relational databases. Relational databases are very common, but relational data models can be tricky to modify. There are other types of data modeling techniques that are best for different cases, such as document databases - which are more flexible - and graph databases - which are suited to particular types of queries.
One key decision for an influence mapping project is how relationship between people and organisations or other type groups are modeled. For a large majority of cases, a simple relational model will be sufficient. Graph data models and databases are more rarely needed, for specific types of queries such as discovering paths in relations or calculating a social graphs' mathematical characteristics - such as centrality or betweenness. There is an extra cost and effort related to using these types of structures because experts are harder to find and the tools are less mature.
Perhaps the most sophisticated and difficult type of data models you can use are based on the semantic web approach. It allows to describe data in a lot of detail and with a lot of flexibility which can be useful to describe complex systems. Using such data models in practice is however cumbersome and only well resourced projects with semantic web experts should attempt to go this route.
Data models should be only as sophisticated as they need to be to answer your project’s questions. It can be tempting to try to capture everything with your data model, but over-engineering your data model will affect how practicable your project is. Adding too many variables and types of entities and relations can result in making the data collection and clean up process much longer, as well as make the development of queries and applications more challenging.
How responsive do you need your queries to be?
Database query times can vary from almost instantaneous to several days, so it’s important to consider performance as part of your project plan. If you’re hoping to make an interactive interface to your data that others will use, for instance, then your queries need to happen without a time lag for it to be useable.
Performance can be affected by your choice of database type will as will the details of your data model. Your choice of data model will also influence whether or not the data and the queries can be distributed on a large number of computers in order to make the queries faster. As performance is controlled by multiple factors you might want to consider discussing this with a data modelling expert before making your choice.
The data you collect and how you’re going to use your database also impacts on performance. Different query types vary in the time needed to return results. You might want to look up a particular data point ("Who was system administrator of this organisation between 2012 and 2014?"), do a text search ("Find all jobs with 'Administrator' in the title") or carry out numerical analysis ("What is the average length of time that a system administrator will stay at her job?"). Similarly, if you're looking at only one company with hundreds of employees it will be quicker than an equivalent search through a data set for an entire country with historical data for tens of years
You can find information about data modelling practices here:
- Exposing the Invisible Decoding Data Guide has a great chapter on structuring data
- The Global Investigative Journalism Network has a good list of data journalism guides
- Agile Data’s Data Modeling 101 This rather technical guide is focussed specifically on data modeling in an agile methodology for app development, but provides a good run-down of data modeling intricacies
Have you applied or developed a practice that you would like to share with the influence mapping community? Edit this post on Github!