Team Spotlight: What I Learned as a First-Year Location Data Analyst

By Kyle Ramos, Data Analyst, Precisely PlaceIQ

Looking back at my first year of being a data analyst at PlaceIQ, the skill I’ve come to value and admire most is the ability to get ‘unstuck’. While this hurdle is something we all try to avoid, it is important to realize that behind even experienced analysts, there is likely a long list of stuck projects. As analysts, we can get stuck in more than one way. We are either technically stuck – lacking the skills to overcome coding errors, or we are conceptually stuck – where our investigative efforts are not answering the question at hand. In this blog, I will share my methods for exploring location data to help fuel your journey in breaking theoretical walls.

Building a process to get unstuck

Because of the highly adaptable datasets here at PlaceIQ, I have found that having technical skills to create metrics opens more avenues to go down in case a roadblock occurs. I did not start this role with strong coding skills, but I have made gains by collaborating with technical product teams like Data Science and Engineering to understand the tools they have created to help wrangle these datasets. Since we often analyze similar datasets, their tools were useful to me. Not all analysts will need to build a technical base like I did, but I believe most can benefit from a get-unstuck contingency plan.

My approach is very simple and requires making a choice from the following three steps: zoom-in, zoom-out, or pivot. I typically follow the steps in this order because each following step will require extra time and extra coding. To explain each concept, I’ll be using PlaceIQ’s Place Intelligence Metrics (PIM) dataset to showcase real world tricks that I’ve used.

Zoom-In

The goal of zooming-in is to understand more about the big picture by looking at its parts. I start this process by looking at the granularity of my data. In PIM, each row represents a metric for a spatial feature over a week, so this is the farthest we can drill down when calculating metrics. Some examples of zooming-in include looking at daily trends after monthly trends, seeing which chains are driving specific metrics for a category, or breaking up the national values into state level data points. I commonly perform the last example, so I’ll show it below.

I was initially tasked with analyzing grocery store trends for the mountain region of the US. Since the states within this region may vary in visitation scale, I needed to investigate which ones greatly influenced the trend seen in Figure 1a. I was most interested in Colorado because it has 48% of the total grocery stores across the region. I start the transformation by taking the most recent snapshot of our basemap and joining to PIM using spatial_id as the key. I can then choose to perform aggregation calculations like total weekly visits by adding up all device counts over each day of the week based on start_date and state. Finally, the output should allow us to see how the visitation changes from state-to-state over continuous weeks.

Weekly visitation to Grocery Stores in the Mountain Region shows a spike in visits occurring in May (Figure 1a). Closer inspection of the states shows not all states followed the overall trend. Colorado was the main reason for the spike in May (Figure 1b).

Zoom-Out

Zooming-out means considering the other metrics that can be calculated within your dataset. I typically choose this path when I feel my analysis is too complex and requires too much supplementary knowledge to understand. I ask myself the question, “is there another metric that can best tell the story?” If the answer is no, Zoom-in. If the answer is yes, then it’s a good time to start considering other metrics. I like PIM since the schema allows for an easy shift from one metric to another. Recently, the ability to shift came in handy when I was tasked to investigate which visitors are more likely to go to a category.

Due to the open-ended nature of this task, I started my analysis with a gut feeling that age may result in interesting trends; however, the initial takeaway was confusing to explain in simple terms. I found the percentage share of visitors over the age of 55 who go to coffee shops is 3% higher than those under the age of 45. I decided to zoom-out since explaining a percentage difference of a percentage may be confusing to grasp. In search of a better metric, I felt calculating the likelihood of visitation can answer the prompt in a clear way. Looking at likelihood, I found a more interesting story comparing incomes. I found that households with less than $50k income are 13% more likely to go casual dining than households that make above $100k. Although both views answered the prompt, the latter story can be understood without external help or additional analyses.

Figure 2: When communicating demographic affinities to different categories, I found calculating likelihood would best express this trend.

Pivot

Of all methods to get unstuck, pivoting is typically the last option since it either requires finding external datasets or an alternative internal one to tackle. Like zooming-out, the goal is to find another metric; however, this metric cannot be calculated with the current dataset. One trick to help speed up the pivoting process is to note which features can be keyed. Using a key can aid in linking to the previous dataset or filtering in the new dataset.

PIM can be a great dataset for pivoting since it can be keyed on spatial_ids. This comes in handy when working with Firehose datasets because the spatial_ids of interest should already be known. Some examples of metrics that can be done in Firehose and not PIM are location level cross-shopping, single day visit analyses, and device level analyses.

Figure 3: Knowing the features that can be used as a key can help when jumping to another dataset.

Getting unstuck can be an overlooked skill since it is typically associated with having inadequate knowledge as a data analyst. This is false. Big data is dynamic and requires exploration to answer new questions. While the methods mentioned here work well with the PIM location data I often use, I encourage you to create or adapt your own process.

Having a plan, applying it to scenarios, and adapting it for future analyses can bring many benefits along your analyst journey; including: building a repertoire of metrics to reference for future analyses, decreasing turnaround time for projects, extracting more in-depth insights, and feeling better prepared for meetings and presentations. If you find this post helpful, feel free to share tricks that have helped you overcome major roadblocks.