Warning: this content is older than 365 days. It may be out of date and no longer relevant.

You Ask, I Answer: Scientific Method for Marketing Data Science?

Jessica asks, “What is most common scientific method to analyze data, so when I (business person) is working with marketing data scientists I can have a intelligent conversation?”

To my knowledge, there is only one scientific method. What matters for marketing data science (and data science in general) is the implementation – particularly at the point where you do your exploratory data analysis. That’s a phase that we skip over far too quickly.

You Ask, I Answer: Scientific Method for Marketing Data Science?

Can’t see anything? Watch it on YouTube here.

Listen to the audio here:

Download the MP3 audio here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.

In today’s episode, Jessica asks, What is the most common scientific method to analyze data so that when I, a business person is working with data scientists, I can have an intelligent conversation.

To my knowledge, there really is only one scientific method, which is you develop a question, you define the problem of the data.

you formulate a hypothesis, you create a test, you run the test, collect the data, analyze the results, you refine your hypothesis or throw it out, and then observe and repeat the process.

Now, all that said, the application of the scientific method is where things differ from traditional science a little bit, not a lot, but a little bit.

Let’s say you’re testing a new vaccine right? For SEO Coronavirus, you would have a question, does this action work? Right? does it create antibodies? You would define the parameters, you would do the formulation, and you would run the test where data science is slightly different is you still have the question you want answered.

But in the problem definition itself, that’s where you’re going to do a lot of what’s called exploratory data analysis.

And that is to understand the problem better to define it better to experiment a little bit, not a lot, but a little bit to analyze the data set itself, if you have it, and to do a lot of refinements to it, cleaning of the data, etc.

so that you can formulate a hypothesis and understand what it is you want to ask and and define the parameters of the test.

Let’s say you want to know Let’s say you want to know the impact of Twitter on your lead generation? That’s a good question, right? What is Twitter’s impact on my lead generation? What data you’re going to need? You’ll need Google Analytics data, probably you’ll need Twitter data.

And you’ll have a hypothesis that you’ll hypothesize that you tweet.

If you tweet more your conversions will go up maybe by a certain amount.

How much is that amount? Do you know? This is where you take that step back into the Define stage of the process and go Okay, let’s look at my Twitter data.

Let’s look at my Google Analytics data.

Is there a mathematical relationship there? Is there even a there there before we set up a test before we create a hypothesis that is testable? Are we even barking up the right tree? So you might run like a regression analysis and see if there is a a mathematical assumption Between the two because without an association, there probably isn’t going to be a causation.

And who would explore your Twitter’s data? Does Twitter give you enough data to build? A good hypothesis? Or conversely, does Twitter give you so much data that you’re not even sure what to test? Right? Is it length of tweet? Is it the day and time of a tweet? Is it a certain emotion or sentiment or tone? That has been working for you images in the post video in the post? What What kinds of stuff do you have access to? And then what kinds of things actually matter? And this is where you would run things like feature selection or feature importance or predictive predictive strength on your Twitter data combined with Google Analytics data.

So there is a fair amount in that defined stage that to our discredit, we tend to gloss over that I would just define success There’s a lot that actually goes into that.

And that it really is exploratory data analysis, which is almost a discipline unto itself, of being able to explore data and understand this is what’s in this thing.

This is what’s in the box, or, and this is where your subject matter expertise really is important.

And your analytical expertise is really important.

What happens when you see an association? Like, yes, it looks like Twitter data has a correlation to Google Analytics conversions.

But then when you run regression tests and things to try and isolate what are the most important features, you come up with nothing, like I just tested a whole bunch of things.

What, what happened here, there’s a there’s an association.

So there should be correlative variables that contribute to it.

But none of these show any kind of statistically valid, predictive strength, what happened and so that that In the experienced marketing data scientists would say okay, what am I missing? I’ve got data, but I’ve got no statistical relevance.

What’s not in the box? What? What else do I need to go and get? And that’s where you’ll find your biggest challenges because it’s tempting to run the test and say, Okay, here’s the top thing, but at the top thing is, you know, a point 08.

And you’re looking for point two, five or better for some of these multiple regression tests, you’d be like, Hmm, what do I do? The very junior or naive data scientist says, I’ll just take the top three, that’s good enough, right? That’s the the algorithms but that’s good enough, but it’s not the case.

That is very rarely ever the case.

Like I’m trying to think of a situation where that is the case and I’m not coming up with anything.

And so for what we want to do, as as As people who want to become experienced marketing data scientists, we have to say, Okay, well, there’s clearly something else that’s missing, there’s a variable that is we don’t have that would glue these two datasets together, or combination of variables.

On the flip side of that, you get things called like a completed variables where there’s something that is mixing the two up and creating stronger signal strength than there actually should be.

That typically happens with highly correlated variables mixed together.

So if you simple example, if it turns out that the length of a tweet is important, and you have the number of characters in the tweet and the number of words in a tweet, and that goes into your, your algorithm to determine strength, that’s going to screw things up because those two are perfectly correlated.

And it’s going to create an an unnecessary signal for the algorithms.

The regression algorithms used to say like, Oh yes, this is invalid.

And contributor to the outcome.

Again, as as a more experienced data scientist, you would look at that and go, Okay, well, we’ve got to get rid of one of the karlitz here, because they both can’t be in there, but they’re gonna throw a wrench into the computation.

So, in Twitter’s case, again, you’d use your social media expertise, Twitter accounts at the character level, if you have 280 characters to work with.

Okay, so if that’s the case, let’s get rid of number of words in the tweet, and just stick with the characters in the tweet.

And that will be a better measure of whether the length of a tweet is relevant to the outcomes that we care about.

So to sum up, there is only one scientific method that I know of.

But it is the implementation that matters the most and how you do it within data science.

Because there’s a lot that goes into defining the problem you will spend a lot of time you should spend a lot of defining the problem.

If you don’t, if you immediately jump into a hypothesis, you immediately jump into running a test.

Chances are, something has gone missing, something has been omitted, that will come back to bite you in the end.

I can virtually guarantee you that something will come back to haunt you and you will not you will not enjoy the process of having to redo the experiment from scratch.

Really good question.

We could spend a whole lot of time on this, but that’s a good starting point.

If you have follow up questions, leave in the comments box below.

Subscribe to the YouTube channel in the newsletter, I’ll talk to you soon take care.

want help solving your company’s data analytics and digital marketing problems.

Visit Trust insights.ai today and let us know how we can help you


You might also enjoy:


Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here


AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!