How To Ensure High-Quality Data
If you cook, you’ve followed a recipe or two. For example, if you’re cooking up a stew, you can fudge a bit on the amounts and maybe even the ingredients. But if you’re baking, precision is key. Skip an ingredient or get the amount wrong and you will have wasted hours and produced an inedible lump. I know, I’m one of those cooks who fudges a lot. But when it comes to baking AND measurement, I’ve learned my lesson: You have to have an accurate recipe and follow each step with precision.
Think of your Measurement Coding Book as the recipe for perfect data. Without it, your chances of success are slim. Over the years, we’ve found that one of the biggest sources of data errors is a bad or incomplete coding document. Whether you decide to outsource human coding or do it in-house, in order to get consistent data you will need to create (or approve if vendor-created) a set of coding instructions to ensure that whoever is doing the reading accurately categories each post or mention.
This involves the following 4 steps:
Step 1: Create and validate your content sources and search string
The search string is a list of all the possible names, brands, products, executives, initiatives, and campaigns that might be influential to your business. It is arguably the key to everything. Get it wrong and you’ll be missing data and/or you’ll end up with dirty data. No matter what, allow plenty of time to create it, test it, validate it, fix it, and test it again. Getting the search string right is not a process that can be rushed.
Step 2: Write, test, and validate a set of coding instructions
The coding instructions should contain all the parameters you’ve agreed to. If you’re coding in Top Tier media outlets, the list of outlets needs to be front and center. If you’re tracking individuals, the complete list should be in the coding instructions. You should also be specific about what should NOT be included.
Does your research include owned media? …paid media? Spell out what should be included. Most importantly, provide examples of EVERYTHING. Every parameter and description MUST include an example. Without examples, coding accuracy goes down by 50% at least.
□ Define your subjects. These are those big bucket that data gets classified into. They might be lines of business, divisions, or departments that you will need to report on. Typically including the priority campaigns, subjects, and initiatives that will occupy most of your time, effort, and energy. Think of these as perennial buckets that media covers and that you will need to report on.
□ Define your themes or topics. These are the priority issues for you for the coming year. Typically they are the battles, campaigns, or issues that are part of the annual strategic plan.
□ Make a list of your key messages. Include no more than 5. If your key messages are long, whittle them down to a simple phrase that is likely to show up in some form in the media.
□ Define what you would consider “desirable” vs. “undesirable” coverage. In other words, what you want to see in print vs. what you don’t. Neutral and balanced is generally considered “desirable.” We recommend using “desirable vs. undesirable” because it simplifies your reporting at the end, and requires less subtlety and interpretation than “positive vs. neutral vs. balanced vs. negative.
□ Make a list of thought leaders, executives, and spokespeople who are important to track.
□ Decide what you’re going to do about visuals. Many of the tracking systems pick up text only, so you may or may not know if a visual is available without clicking through on a URL. Are visuals important? If so, allow enough time to click thru on a URL, or hire a vendor who provides visuals. Define what a desirable visual vs undesirable visual might be. As much as we all would love front page photos of our brand, you may not want it on the side of an ISIS truck.
□ List anything else that is important to track such as the type of conversation, influence scores, number of shares, etc.
Step 3: Test
Give these lists to the designated reader or vendor and have them test read 50 articles. Correct and revise your sources, search string, and instructions as necessary.
Step 4: Test again
Revise as needed until the coding reflects what is important to your organization.
Originally published at painepublishing.com on March 21, 2016.