White women find working with me polarizing. I know because I asked.

(This is a post primarily about the results of an experiment, but I did build a tool as part of it that you can use to gather behavioral feedback from former coworkers; you can go directly to WorkWithMeAgain.com to use it for free without reading about the journey or my data.)

It started, as so many of my projects do, with a conversation on Twitter. I proposed a public Glassdoor-like system with a single rating: would you be willing to work with this person again? Slice that by demographics and we might have a tool worth building to increase inclusion in the workplace.

After much healthy debate about the risks (it could be used as a harassment tool, for example), much of the value seemed to stem from the collection of feedback itself. Shifting away from the goal of a public system for discovering bad actors, it is worth asking: does the average person even know whether former coworkers would want to work with them again?

I certainly don’t and especially not in any systematic way; despite being much-touted by thought leaders in the management space, organized feedback collection is relatively rare. For years, I’ve had a link at the bottom of my email that allowed people to leave me anonymous feedback via Google Form and I’ve received plenty of responses, some helpful and some less so (apparently, someone doesn’t like my cowboy boots). But I never collected demographic data or any concrete behavioral metrics, like willingness to work together again.

So I started building. I iterated through several versions of both the survey questions and the analysis before arriving at the results I’ll talk about today. The dominant behavioral measure is a simple 1–5 scale: “Given the opportunity, would you actively avoid or seek out working with me again?” With that, I asked gender and ethnic identity, age when we worked together, reporting relationship, and recency of the relationship.

To avoid as much bias as possible, the final survey was sent out indiscriminately to ~800 people I’ve worked with over my ~15 year career. The sampling wasn’t perfect, because people have moved between companies, I’m not connected to everyone I ever worked with, and many people who have worked with me don’t have LinkedIn profiles (due to socioeconomic skews of who is on LinkedIn). But this whole project is a lesson in not letting the perfect be the enemy of the good; it is easy to find reasons that feedback will be incomplete, but that isn’t a reason to not get feedback at all.

In all, I received ~70 completed surveys, which seems extraordinarily high given what we know about average survey response rates. Since I phrased the survey request as an experiment, the results of which I was committed to publishing, there may have been some increased promoting pressure but I’d like to believe that we might actually be on to something here; people want to give feedback, if we choose to hear it.

Before we get into the result that drove the title of this post, let’s start with resolving one important concern that came up in the Twitter discussion: that many people would choose not to add demographics for fear of being identified. This didn’t appear to be the case, as only ~10% chose not to volunteer full demographics. Assuming that response rates were high enough, it is thus likely that meaningful demographic filters can be applied.

Now the results at a high level, with the boring stuff first. If you want the nitty gritty, I’ve linked to a copy of the report so you can see what you’ll actually get if you do this yourself using WorkWithMeAgain.com, although I have redacted the free entry portion to preserve the anonymity of respondents. I also hadn’t yet added the recency question (which was the great suggestion of Leigh Honeywell) in the version of the survey I sent out, so there is no data for that section.

The basic finding is that statistically speaking, there were no significant differences in the averages within demographic groups. Men/women/non-binary people and white/non-white people were equally likely to want to work with me. And the averages were all above three on the scale, meaning generally all of those groups would seek out working with me again.

Does that mean I’m free of bias? Of course not; as a white male, I absolutely have racist and sexist associations that express themselves as behaviors, despite my intentions. What it does mean is that at least in this sample, these biases were not large enough to meaningfully affect any particular demographic subsegment, which is a good thing.

There are some interesting correlations that are neither good nor bad. For example, the older someone was when they worked with me, the more likely they are to want to work with me again, which means I may need to change how I relate to younger people in the workplace. I can come across as condescending, so this is a known problem that I actively work on, and that data backs it up.

Where the data starts to get interesting is when we look at standard deviations. So rather than how much would people seek me out generally, how much disagreement was there within the group about that behavior.

I am, unsurprisingly, a fairly polarizing coworker. I’m outspoken on social issues, have a distinctive personality and work style, and frequently clash with others who I perceive as acting against the group interest. I’m not unaware of this, although I do vary at times in how much I see it as a strength, weakness, or simple fact.

This polarization, however, is not evenly distributed. Even though men and women/non-binary people don’t differ in their average desire to work together again, they do differ in their polarization: women have much stronger feelings about avoiding or seeking me out, as do people I managed and people who worked with me across teams. In my case, because I have enough data to cross gender and ethnicity, the finding is really around white women: men and non-white women consistently would seek out working with me, while white women are basically either 1s or 5s.

This is in someways rather puzzling. For example, I once lead a customer service team of ~100 people who were almost all non-white women and subsequently outsourced the function, resulting in a significant layoff. While I remain convinced that it was the right decision that was made in the best interest of our customers, all of those people would be entirely justified in avoiding working with me again. So I would have expected a potential main effect of ethnic identity or at least an interaction with gender.

I’ll confess, I don’t know what the varying feedback from white women means yet. And that’s a fine thing: survey data like this rarely gives us a conclusive answer. Instead, good quantitative analysis clues us in to where to do qualitative analysis; data tells us what, not why. For me, the survey is a jumping off point to conversation that I can use to improve and I intend to seek out further conversation, particularly with white women who I’ve worked with before.

Feedback isn’t a silver bullet for changing our behavior. But it can be a start. Personally, I’ve committed to publishing this data and to following up with the 800 people in the initial sample, as well as periodically adding new co-workers with each new job. As part of that, I’ll also resurface the link to the anonymous feedback form I already have setup; for any of the 10 people who would prefer not to work with me again, if you are willing, I’d deeply appreciate you using that to give me feedback on what I could have done differently to change your experience with me. Finally, readers of the writeup can use the same form to give suggestions on actions I should take or email me at matt@mattwallaert.com if you don’t need anonymity. I want to do better, I’m committed to doing better, and I hope when I one day rerun this experiment, the results will be different.

If you’re committed to doing the same kind of work, the tools I used are now publicly available at WorkWithMeAgain.com. It will allow you to copy the Google Sheet that contains all the calculations, as well as instructions on how to launch the survey to your former coworkers. I am absolutely convinced that putting in the effort to gather meaningful behavioral feedback can be a key component in how we change our individual behaviors, as the collective culture is simply made up of those individual behaviors; if enough of us do this, we can live in a substantially better world.

Side Note: As usual, this project required collaborators: frontend coding from Grayson Null and design by Patricio Hunterkhozner. Plus feedback from a WhatsApp group full of folks working toward inclusion and 70 people who were willing to share their thoughts about working with me again. It has been a good year for WorkWorthDoing projects (Mediocre White Men and Project 4255, with another on the way soon) but they don’t happen without a bunch of people willing to work at reduced rates and in suboptimal circumstances.

Posting my survey results, online and unfiltered, is making me nervous. And that’s a novel feeling, since I usually don’t really feel nervous in circumstances that don’t involve heights (I really don’t like heights). That makes this a rare brave moment, because if I don’t normally feel the feelings, I’m not actually being brave: it is easy to do something with low inhibiting pressure. There is no concrete outcome I am afraid of…but I’m certainly afraid.