LPRC - CrimeScience – The Weekly R2P Review – Episode 227 Ft. Sam Yeung, PhD

Episode Date: January 15, 2026

In this episode of LPRC CrimeScience Podcast – The Weekly R2P Review, we delve into a groundbreaking guard study featuring insights from Dr. Sam Yeung. Dr. Yeung shares key findings that highlight the implications of guard presence on crime rates and overall safety. Listen in to the newest R2P Report!

Hi everyone and welcome to crime science. In this podcast, we explore the science of crime and the practical application of this science for loss prevention and asset protection practitioners, as well as other professionals. Welcome back, crime science guests. Thank you so much for tuning in. I am your host, Alex Palomar, research project coordinator at the LPRC, and I appreciate all of our listeners. Today on the podcast, we have Dr. Siamen, Sam Young, who is a research scientist at the LPRC. Sam, I'm so happy to have you today. Can you please tell me a little bit about yourself and your background?

Sure, thanks Alex for having me. My name is Cam Long Young, and I go by Sam. My background is really a mix of building things and study how people think. I start out as a civil engineering undergrad at Cornell, but later in my senior year, I become interested in human behavior. That's why I pursued cognitive psychology PhD from Iowa State University. For a couple of years, I was doing research in the academia, but later I picked up data science and got a micrombaster from Georgia Tech.

In the last couple of years, I've been, I would say, a research data scientist at LPRC, where I do a big data project, a randomized control trial, simulation of active shooting incident to study how how to better design or what factors can minimize injuries and fatalities in active shooting situation in the retail environment. Yeah, that's very fascinating. And I'm really excited to go into your R2P today, the Protis Guard study. So we can just start off by having you recap the article for those who may not have had the chance to read it yet, more so what the recent.

was, why it was conducted, how it was conducted, and so forth. Yeah, sure. The essence of the article and the study is that we wanted to answer one big questions. Do security guards actually stop crime in stores? It sounds like an obvious, yes, but guards are expensive, and stores owner want to see prove that they are worth the money. So the problem is that stores usually only hire guards when they are already in trouble, meaning that the crime level there is already intermediate to serious.

So if you want to look at the numbers, a store with a guard might have more theft than acquire store with a one, not because the guard is doing a bad job, but because it starts with a higher level of crime. So in order to take that into consideration, we look at two years of data from 4,000 stores across the US, and we track every time a guard was there and every time something was stolen to cut as an incident. So instead of comparing different stores, we compare each store to itself.

So essentially using its own information as a control. Look at how specific stores did on days we've got versus days without one. Okay, and what were the main findings of that? The main takeaway is that guards definitely help. we found that whenever a guard was added, the number of clients ran down across every type of store that we study and the result is statistically. And I think meaningfully significant depends on how the retailer interpret the information. For example, in the big stores, we saw the biggest wing for every increased number of guards per one guard. I think it's a 0.33 decrease in daily incident.

And for a specialty store, the number is a little bit smaller. It's around at a 0.1. Similar magnitude of reduction is also seen in other type of store. And in the beauty and health store, we see that having a guard as emergency short-term help to save the store $90 per day, and that is the only significant reduction of retail loss rather than just incident. And what do you think was the reason for the difference in incident reduction in big box retailers versus the specialty versus the health and beauty?

I think the major driver is that a big box store by definition has a big box. footprint compared to a specialty store and the other type of stores. So that is one main possibility, I would say that's why there's a larger perceived reduction. Okay, that makes sense. And then you've mentioned a few times in your report that the guard coverage was neither randomized nor systematically assigned. How did you account for the non-randomization of guard? randomization of guard deployment. And can you explain that multi-level modeling a bit?

Yeah, sure. So to make it fair, we had to use a special statistical method called multi-level modeling. Just think of judging a track coach, you wouldn't just look at who has the fastest runner. You look at how much faster each runner goes after the coach start training them. So some runners are naturally faster than others,

just like some students. stores are in tougher neighborhoods than others. So the statistical method allows us to level the playing through across those, just as in the coach cases, the runners are compared with the same similar type of references. So essentially, the method helped us to lay the mathematical foundation to have a fair comparison. even different stores have different crime level to begin with. And if you were to repeat this study, would you want it to be more randomized? And if so, how would you be able to do that?

In the perfect way, we would like to have a complete randomized control trial that we draw the stores from some random sample algorithm, maybe a random number generator, and then put the stores there. and then start them, say only pick the first 50 stores out of 5,000 store, for example. But it will be counterintuitive to the business metrics or judgment, because you also want to take into consideration of the fact that you want your ROI to have the biggest impact. And the biggest impact usually come from the store with the highest crime. So to do it again, I would suggest to have partial or stratify random sample, meaning that out of the maybe the top third stores with high retail crime, I only sample, I have sample from that strategy as far as a sample from the

middle, one third with the crime level, and then, the safest store. That way we can have a more kind of random sampling procedure that is better than the current one, but it is not completely random sample. I hope I'm making sense here. You're definitely making sense. And now just getting done with the holidays, you also mentioned that seasonal fluctuations affect the results of this study. Does that go for just the holiday season, or is it the summer, the spring, how does that affect it? That's a good question. Seasonal is definitely an issue here.

And depending on the level of death, we want to look into the effect there. For example, there has been quite consensus findings about crimes are more pronounced in the summer. days than in the winter and more so after work hours than in during work hours just because people go out more frequently in summer than in winter. So at the seasonal level we can look at that and try to control it statistically in the model. Same for the hour, but we just do not have enough data resolution, meaning that we do not have the number of incident counts per hour or per afternoon. or per night. Otherwise, we could go deeper into that realm of investigation. Okay. Yeah, that definitely makes sense. And then you had mentioned again in the report that this study took about two years to complete. In the future, if you were to do another version of this study, is two years kind of that sweet spot or would you want to do it longer, shorter?

Um, um, this is a very good question because things are changing very rapidly in our society. And at the time that we're speaking, there may be protests happening in Minnesota. So, uh, so the longer time frame would be always better in most cases, but also we need to take into account historical events, like the ones, or that is happening in Minneapolis or Minnesota in general. So as long as we have the data that can take into account, a longer time horizon is usually better. And then what can industry leaders take away from this study?

And how can practitioners apply it to their program? So I think the key takeaway is that guards are effective in terms of reducing the number of crime incidents that are reported from the case management system. But the key takeaway should be that retail industry leaders should take that into consideration and to see if the cost of guards compared to the benefit they can get from it would be worth it. Point-free incident by sound like a small number, but if people feel safer in general,

they may be more inclined or they may be happier there. So the customer experience are also affected. And because of due to their perceived safety there, they might be more inclined to go there. So there are a lot of other factors that the industry leader need to take into consideration. And it looks like from the report as well that you had mentioned full-time guards and only emergency services guards made an impact. So it wouldn't have to be a full-time guard. that's one of the takeaways, right?

It's just any guard presence. I would be cautious about the interpretation. It's not that you're wrong, but I think the majority of the data shows that they are from a temporary service, rather than long-term service, which makes sense in the retail industry, because unless there's a need, I wouldn't just spend the money to hire full-time guard all the time. Okay, yeah, and that's a great,

Clarification. Thank you for that. Well, thank you, Sam, for taking time with us to go over your report. It was definitely a good one, and it was due time for us to get on the podcast together. So I appreciate your time. Thank you very much for having me. And those who are listening, thank you again for sticking around. If you are interested in looking through Sam's entire report, you can visit our Knowledge Center and search his Proto's Guard study. Thanks again, everyone. See you on the next one. Thanks for listening to the Crime Science Podcast, presented by the Loss Prevention Research Council. If you enjoyed today's episode, you can find more crime science episodes and valuable information at LPRsearch.org. The content provided in the Crime Science podcast is for informational purposes only and is not a substitute for legal, financial, or other advice. Views expressed by guests of the Crime Science Podcast are those of the online.

authors and do not reflect the opinions or positions of the Loss Prevention Research Council.