WEBVTT 1 00:00:03.720 --> 00:00:16.610 Danyi Li: Welcome to the Tobacco Online Policy Seminar, TOPS. Thank you for joining us today. I'm Danny Lee, a PhD candidate in Health Behavior Research at CAICS School of Medicine, University of Southern California. 2 00:00:17.000 --> 00:00:32.149 Danyi Li: TOPS is organized by Mike Pascoe at the University of Missouri, Sun Shan at the Ohio State University, Michael Darden at Johns Hopkins University, James Hardman-Boyce at the University of Massachusetts Amherst. 3 00:00:32.479 --> 00:00:36.550 Danyi Li: and Justin White at Boston University. 4 00:00:36.950 --> 00:00:41.520 Danyi Li: The seminar will be one hour with questions from the moderators and the discussant. 5 00:00:41.630 --> 00:00:51.129 Danyi Li: The audience may post questions and comments in the Q&A panel, and the moderator will draw from these questions and comments in conversation with the presenter. 6 00:00:51.420 --> 00:00:58.190 Danyi Li: Please review the guidelines on tobaccopolicy.org for acceptable questions. 7 00:00:58.600 --> 00:01:03.290 Danyi Li: Please keep the questions professional and related to the research being discussed. 8 00:01:03.730 --> 00:01:11.510 Danyi Li: Questions that meet the seminar series guidelines will be shared with the presenter afterwards, even if they were not read aloud. 9 00:01:11.990 --> 00:01:14.560 Danyi Li: Your questions are very much appreciated. 10 00:01:15.050 --> 00:01:24.809 Danyi Li: This presentation is being video recorded and will be made available along with presentation slides on the TOPS website, tobaccopolicy.org. 11 00:01:25.530 --> 00:01:33.259 Danyi Li: I will turn the presentation over to today's moderator, Michael Darden at Johns Hopkins University, to introduce our speaker. 12 00:01:35.290 --> 00:01:52.240 Michael Darden: Thank you. So today we're going to continue our summer 2026 season with a single paper presentation by Andreas Stoller entitled Impact of Tobacco Advertising Restrictions in A Quasi-Experimental Study on the Effect of Billboard Bans on Smoking. 13 00:01:52.240 --> 00:01:57.639 Michael Darden: The presentation was selected via a competitive review process by submission through the TOPS website. 14 00:01:57.860 --> 00:02:14.760 Michael Darden: Andreas Schoeller's research assesses the effectiveness of prevention policies such as tobacco taxes, tobacco billboard bans, and regulations on e-cigarettes and heated tobacco products. He's an applied microeconomist using state-of-the-art econometric methods for policy evaluation with an emphasis 15 00:02:14.760 --> 00:02:28.530 Michael Darden: on causal inference and causal machine learning. He previously worked on social security and labor market interventions at the Swiss Federal Statistical Office and the University of Basel. So, Andreas, thank you for presenting us for today. 16 00:02:31.820 --> 00:02:38.509 Andreas Stoller: Hello, thanks a lot for the nice introduction. I will share my screen. 17 00:02:38.910 --> 00:02:39.930 Andreas Stoller: Up. 18 00:02:44.520 --> 00:02:52.189 Andreas Stoller: Now you should be all able to see the screen. Please tell me if not, I cannot see you anymore. So… 19 00:02:54.470 --> 00:02:54.950 Michael Darden: Thank you. 20 00:02:54.950 --> 00:03:02.090 Andreas Stoller: Thank you all for having me. I really feel honored to present here at the Tobacco Online Policy Seminar. 21 00:03:02.200 --> 00:03:10.800 Andreas Stoller: I will proceed to the disclosures slide. So I didn't receive any funding. 22 00:03:10.920 --> 00:03:14.950 Andreas Stoller: From tobacco, or pharmaceutical companies. 23 00:03:16.100 --> 00:03:34.070 Andreas Stoller: What I received for funding for this work, and which is also the only funding I received over the past 10 years, is funding from the Swiss Tobacco Prevention Fund, where I analyzed tobacco prevention policies in Switzerland and Europe. 24 00:03:36.600 --> 00:03:55.560 Andreas Stoller: I think with this audience, I don't have to say it, but tobacco remains a leading preventable risk factor. Despite the declining smoking rates, we have about 8 million deaths worldwide attributed to tobacco. 25 00:03:55.950 --> 00:04:10.689 Andreas Stoller: Europe has still the second highest smoking rate, worldwide, with about, 800,000 deaths. So, also in this region, this remains a very important policy question. 26 00:04:10.950 --> 00:04:17.129 Andreas Stoller: So, this to say, smoking remains a major public health challenge. 27 00:04:17.829 --> 00:04:33.870 Andreas Stoller: Of course, tobacco prevention policy is a big topic in health economics. Many papers focus on analyzing tobacco taxation, also the smoking bans in public areas, sales bans for minors. 28 00:04:33.870 --> 00:04:41.260 Andreas Stoller: and advertising bans. Then there are a few interventions or prevention. 29 00:04:41.600 --> 00:04:56.459 Andreas Stoller: techniques that are maybe a bit less studied, the information campaigns or school programs, quitlines. Maybe they lack a bit in economic aspects, so they are a bit less represented in the health economics literature. 30 00:04:56.600 --> 00:05:10.159 Andreas Stoller: Interestingly, the causal studies, so the studies exploiting a quasi-experimental design, mainly focus on taxation and smoking bans. Maybe there are a few on sales bans, but 31 00:05:10.170 --> 00:05:23.700 Andreas Stoller: Then, the causal evidence for advertising restrictions sadly remains limited, especially for, tobacco advertising. I am aware of a few, good, quasi-experimental studies 32 00:05:23.750 --> 00:05:42.480 Andreas Stoller: On e-cigarette advertising or e-cigarette advertising bans, however, for tobacco advertising, I'm not aware of such studies and except for mine, of course, which fills that gap. If anyone knows about such studies, please tell me about it. 33 00:05:43.630 --> 00:06:01.540 Andreas Stoller: I further want to motivate with this paper by thinking about why a company would even want to advertise their brand. There is one main argument that is typically put forward by the industry, namely they just advertise to increase their own market share. 34 00:06:01.740 --> 00:06:07.539 Andreas Stoller: Attracting smokers who already smoke to just smoke their brand instead of another. 35 00:06:07.690 --> 00:06:18.650 Andreas Stoller: While the total demand for cigarettes or tobacco products remains the same. So the argument is that consumers would just shift from one brand to the other. 36 00:06:18.890 --> 00:06:24.540 Andreas Stoller: However, there is also this market expansion idea that 37 00:06:24.990 --> 00:06:43.039 Andreas Stoller: advertising actually persuades people and initiates non-smokers to smoke, or increases the consumption among smokers. And it's mainly this second point that I'm interested in. Can one show that advertising increases consumption? 38 00:06:43.190 --> 00:06:50.670 Andreas Stoller: Or the other way around here, like, do advertising bans reduce consumption, and 39 00:06:50.790 --> 00:06:56.779 Andreas Stoller: Thus would justify regulating tobacco advertising for public health reasons. 40 00:06:56.910 --> 00:07:08.999 Andreas Stoller: This is not like an industrial organization paper, it's really about showing the effect of this policy, and to see if such regulation is warranted or not. 41 00:07:10.270 --> 00:07:25.390 Andreas Stoller: So, the main contribution of this paper is that I estimate the effect of cantonal billboard bans, like cantons are sub-regions of Switzerland, as you will see in a moment, and their effect on smoking. 42 00:07:26.400 --> 00:07:43.559 Andreas Stoller: I provide causal evidence using individual-level data, and the findings will support the market expansion hypothesis. So, actually showing advertising restrictions can be an important part of a comprehensive tobacco prevention framework. 43 00:07:46.030 --> 00:07:54.100 Andreas Stoller: Before I proceed to the data, I want to give you a quick overview on the system that we are 44 00:07:54.780 --> 00:07:56.490 Andreas Stoller: analyzing here. 45 00:07:56.580 --> 00:08:00.490 Andreas Stoller: It's about Switzerland and but yeah. 46 00:08:00.510 --> 00:08:16.209 Andreas Stoller: policies enacted by sub-regions of Switzerland. So, we have three main levels of government. Namely, we have the Copenhagen Federation, which is the national level. Then we have 26 sub-regions, which are called the cantons. 47 00:08:16.210 --> 00:08:23.270 Andreas Stoller: And then we have many municipalities at the lowest administrative level or the villages. 48 00:08:24.110 --> 00:08:28.950 Andreas Stoller: Tobacco prevention policies can be enacted at all these different levels. 49 00:08:29.100 --> 00:08:39.339 Andreas Stoller: But it's mainly at the national level, or then this cantonal level, which also then allows us to, use variation in… 50 00:08:39.700 --> 00:08:46.020 Andreas Stoller: tobacco prevention policy across these cantons, which I will effectively do. 51 00:08:48.000 --> 00:08:55.759 Andreas Stoller: The tobacco politics, are such that at the national level, we have the tobacco taxes. There is no 52 00:08:56.000 --> 00:08:58.280 Andreas Stoller: tobacco tax at the lower level. 53 00:08:59.120 --> 00:09:03.429 Andreas Stoller: We have health warnings and the cigarette packages at the national level. 54 00:09:03.700 --> 00:09:10.389 Andreas Stoller: also TV and radio advertising bands, and a few other, like… basic regulations. 55 00:09:11.050 --> 00:09:16.680 Andreas Stoller: Then at the… Middle level, at the cantonal level, we have 56 00:09:16.820 --> 00:09:30.550 Andreas Stoller: billboard bans, so some form of partial advertising restriction, sales bans for minors, and smoking bans, which were mostly enacted at this cantonal level. And typically, it's like. 57 00:09:30.580 --> 00:09:45.230 Andreas Stoller: some cantons first introducing it, then others learning from them, introducing very similar policies at a later point in time, and eventually, after many years, it might become a national level. So, there is some 58 00:09:45.450 --> 00:09:51.560 Andreas Stoller: A national policy, so there is some sort of learning, going on. 59 00:09:51.970 --> 00:10:08.059 Andreas Stoller: And then, of course, we have just interventions that are happening, like information campaigns and other, like, temporarily constrained interventions by either governmental or non-governmental actors. 60 00:10:08.670 --> 00:10:25.149 Andreas Stoller: So just to give you a bit an idea about all the different tobacco prevention policy that can happen in Switzerland. Here we have a map of Switzerland and the introduction of these billboard bands graphically represented. 61 00:10:25.920 --> 00:10:29.609 Andreas Stoller: About the billboard ban, I want to say that 62 00:10:29.930 --> 00:10:39.050 Andreas Stoller: It concerns only billboards that are in public spaces, or that are on private ground, but visible from public spaces. 63 00:10:39.160 --> 00:10:45.030 Andreas Stoller: So, typically what you would see at the side of a street or a highway. 64 00:10:45.190 --> 00:10:49.960 Andreas Stoller: So something that you can quite frequently encounter. 65 00:10:51.520 --> 00:11:07.079 Andreas Stoller: The very first canton introducing this type of ban was the canton of Basel, a very small Swiss-German canton in 1997. The canton of Geneva introduced it in the year 2000. 66 00:11:07.080 --> 00:11:14.109 Andreas Stoller: Now, these were a bit the very first ones that introduced this type of ban, and then, over the years, until… 67 00:11:14.140 --> 00:11:17.890 Andreas Stoller: Like, mostly between 2005 and 2010, 68 00:11:18.060 --> 00:11:21.429 Andreas Stoller: Many of the other cantons followed them. 69 00:11:21.790 --> 00:11:35.069 Andreas Stoller: So, what we have here is really a perfect quasi-experimental design. We have some parts of the population exposed to this type of billboard ban, and another part that isn't exposed to this billboard ban. 70 00:11:35.500 --> 00:11:48.029 Andreas Stoller: And an important thing to remember here, we don't have much social media yet, and billboards were actually the main advertising channel at that time. So, by, 71 00:11:48.410 --> 00:11:57.179 Andreas Stoller: By about 10 times, it was more than all the other media channels that were around at that time. 72 00:11:57.370 --> 00:11:59.680 Andreas Stoller: Due to these bands, then the… 73 00:12:00.430 --> 00:12:05.959 Andreas Stoller: Advertisement expenditures reduced heavily for tobacco billboards. 74 00:12:06.070 --> 00:12:14.179 Andreas Stoller: And interestingly, there was no substitution towards other channels, or almost no. For newspapers, the… 75 00:12:14.510 --> 00:12:33.249 Andreas Stoller: advertising expenditures of the industry maybe doubled or tripled, but because the level of newspaper or print media was so low, it hardly substitutes any of the expenditures that were made for tobacco billboards. 76 00:12:36.760 --> 00:12:44.589 Andreas Stoller: So, so much about the policy, now more about the data that I will exploit in this analysis. 77 00:12:44.710 --> 00:12:53.440 Andreas Stoller: I use the Swiss Health Survey data that is given by the Federal Statistical Office. This is a 78 00:12:53.640 --> 00:12:59.410 Andreas Stoller: These are cross-sectional waves that are given every 5 years. 79 00:12:59.710 --> 00:13:09.850 Andreas Stoller: The interesting part about these cross-sectional waves is that we know of each individual that was surveyed when they started and stopped smoking, so… 80 00:13:09.850 --> 00:13:21.520 Andreas Stoller: This allows to reconstruct the annual smoking status of all these individuals. So we actually have the smoking history of all these individuals. 81 00:13:23.740 --> 00:13:37.260 Andreas Stoller: With this reconstruction of the annual smoking rate, we sadly use some information, namely the time variant information cannot be retained. 82 00:13:37.260 --> 00:13:49.039 Andreas Stoller: So we can just track the outcome, namely the smoking status of these individuals, if they smoke or not, their gender, and their age. I mean, the age we can just 83 00:13:49.200 --> 00:13:51.000 Andreas Stoller: calculate. 84 00:13:51.820 --> 00:13:59.540 Andreas Stoller: So, with this, we end up with about 1 million observations between the years 93 and 2017. 85 00:14:01.030 --> 00:14:14.219 Andreas Stoller: Just to make the reconstructed panel more credible, here I plot once the smoking rate using the original data where we just have data every five years. 86 00:14:14.220 --> 00:14:22.789 Andreas Stoller: and once using the reconstructed panel. And what you see is that the smoking rate is always… is about the same in both. 87 00:14:24.060 --> 00:14:33.840 Andreas Stoller: in both datasets. So, structurally, there doesn't seem to be a big change due to this manipulation. So, the… 88 00:14:34.100 --> 00:14:48.399 Andreas Stoller: One might… may call it recall bias, isn't that… doesn't seem to impact much, the… the smoking rate on average. Like, the… the fact that people might not remember that well when they exactly stopped smoking. 89 00:14:49.470 --> 00:14:50.990 Andreas Stoller: Or started smoking. 90 00:14:51.610 --> 00:14:59.760 Andreas Stoller: So, yeah, I think at this point, we will now go to the discussion, so… 91 00:14:59.760 --> 00:15:16.370 Michael Darden: Thank you, Andres. That's great. So our discussion today is going to be Christian Saenz, a postdoctoral associate from Yale University School of Public Health. His research examines substance use, health behaviors, and mortality. So Christian, hello. 92 00:15:18.550 --> 00:15:26.760 Christian Saenz: Okay, well, thank you very much. Having read this paper, I thought it was well written. The analysis was. 93 00:15:26.910 --> 00:15:37.110 Christian Saenz: thorough and appropriate, and, the research question is obviously an interesting one. And as Andreas notes in his paper. 94 00:15:37.240 --> 00:15:40.759 Christian Saenz: A lot of the prior evidence may not, 95 00:15:41.080 --> 00:15:44.209 Christian Saenz: be the highest quality in terms of 96 00:15:44.330 --> 00:15:51.290 Christian Saenz: a causal interpretation of of how these billboard bands may impact 97 00:15:51.390 --> 00:16:07.459 Christian Saenz: smoking behaviors, and so I do think that this paper is a unique contribution. I do have a couple of minor points, and this might be putting the cart before the horse. 98 00:16:07.560 --> 00:16:14.090 Christian Saenz: But, as it relates to your estimation strategy, so, in… 99 00:16:14.270 --> 00:16:23.559 Christian Saenz: this country, you've mentioned canons or provinces, and I believe there's only 25 or 26. 100 00:16:23.680 --> 00:16:27.320 Christian Saenz: And so that could potentially be. 101 00:16:27.460 --> 00:16:44.390 Christian Saenz: a small cluster problem if we don't have enough geographic clusters, and that might impact our statistical inference. And so, could you perhaps preview if that is a concern or not, and how you 102 00:16:44.390 --> 00:16:53.170 Christian Saenz: Address the, 26… provinces in your empirical strategy. 103 00:16:56.900 --> 00:17:00.970 Andreas Stoller: Should I respond, like, directly, or do you want to ask all the… 104 00:17:01.560 --> 00:17:04.560 Michael Darden: Yeah, you can go one for one. So go ahead. Yes, sure. 105 00:17:04.730 --> 00:17:22.520 Andreas Stoller: Yeah, thank you for raising this question, and I, actually, I do not cluster the standard errors at the cantonal level, which might be not that intuitive, because typically in the textbooks, one is advised to cluster at the level of treatment, and 106 00:17:22.520 --> 00:17:28.889 Andreas Stoller: The level of treatment is the cantons, or, yeah, the provinces, as you say, and there are only 26 of them. 107 00:17:30.700 --> 00:17:35.870 Andreas Stoller: Here, I decided to cluster at the individual level, because, I mean. 108 00:17:36.380 --> 00:17:48.460 Andreas Stoller: from my understanding, clustering isn't about confounding. It's more about the observations, not being independent. And when I think about 109 00:17:48.690 --> 00:18:00.320 Andreas Stoller: the level of clustering were about independence and not about treatment administration and assignment and confounding, then I would argue that 110 00:18:00.900 --> 00:18:17.759 Andreas Stoller: Like as we have here a panel, the past smoking of the individual or like just the smoking status of the individual is the most like dependent, dependent part of this whole regression. 111 00:18:18.270 --> 00:18:30.940 Andreas Stoller: And that's why, I mean, I cluster at the individual level, and I have about 90,000 individuals where there is no problem with small clusters. 112 00:18:31.080 --> 00:18:31.859 Andreas Stoller: You know. 113 00:18:32.080 --> 00:18:36.090 Andreas Stoller: But I know it's not that, yeah… Not that, 114 00:18:36.410 --> 00:18:41.359 Andreas Stoller: Popular, but it's just another way of arguing for it. 115 00:18:42.940 --> 00:18:51.060 Christian Saenz: Yeah, that's a fair point. I do think a worthwhile robustness check could be done. 116 00:18:51.200 --> 00:18:58.239 Christian Saenz: an alternative type of clustering of standard errors at the Canton level. 117 00:18:58.580 --> 00:19:14.139 Christian Saenz: and perhaps using like a wild cluster bootstrap or something, and just showing that your the statistical significance and your inference is unaffected. I think that would be 118 00:19:14.230 --> 00:19:27.750 Christian Saenz: a worthwhile and relatively simple robustness check that could maybe make my comment moot. 119 00:19:27.750 --> 00:19:28.709 Andreas Stoller: Yeah, I agree. 120 00:19:29.310 --> 00:19:38.620 Christian Saenz: I guess one other question, and again, you can defer this until after you discuss your estimation strategy. 121 00:19:38.720 --> 00:19:43.890 Christian Saenz: So the data we're using, you construct, 122 00:19:44.160 --> 00:19:48.020 Christian Saenz: A panel essentially of smoking rates over time. 123 00:19:48.170 --> 00:19:57.510 Christian Saenz: But this is a cross-sectional data, right? And so it. 124 00:19:57.790 --> 00:20:01.729 Christian Saenz: How do we think about, is the data representative? 125 00:20:02.020 --> 00:20:21.380 Christian Saenz: Does it have survey weights where we can weight our analysis in order to make the individuals that we're studying representative of the overall national population or the Canton specific population? 126 00:20:21.560 --> 00:20:27.429 Christian Saenz: I assume that currently you don't use survey weights, or perhaps you do, but if you could 127 00:20:27.660 --> 00:20:37.550 Christian Saenz: clarify what you think the right approach is for that. And that was my only other comment for the moment. 128 00:20:38.500 --> 00:20:55.499 Andreas Stoller: Many thanks for raising the question, and I present often in front of, like, economics and health science publics, and this seems to be really a health science question. They often apply the weights, while in economics it seems to be a bit 129 00:20:55.520 --> 00:21:12.579 Andreas Stoller: something that we do less, or maybe I'm not aware of it. As… I mean, here the main point is that I care about internal validity, showing an effect in the sample, but I see the interest of applying weights. I haven't done that yet, and 130 00:21:12.780 --> 00:21:18.149 Andreas Stoller: It could be, yeah, maybe it's not even a robustness check, but… 131 00:21:18.360 --> 00:21:23.100 Andreas Stoller: Yeah, it could be an interesting addition to see how… 132 00:21:23.280 --> 00:21:39.339 Andreas Stoller: if the effect would have actually be present at the country level of Switzerland. And there are survey weights, but of course they relate to the cross section, so maybe… 133 00:21:39.550 --> 00:21:43.259 Andreas Stoller: I would have to calculate rates myself, too. 134 00:21:43.820 --> 00:21:49.760 Andreas Stoller: Make the results representative, at least by age and gender. 135 00:21:50.260 --> 00:21:58.129 Andreas Stoller: So yeah, it's certainly a very valid addition, yeah, for having more external validity. 136 00:21:59.540 --> 00:22:09.380 Michael Darden: One quick question from the chat. Can you? And maybe you'll get to this. But do you know anything about like the market for billboards themselves? So the. 137 00:22:09.380 --> 00:22:11.439 Andreas Stoller: That's our big one. 138 00:22:11.440 --> 00:22:21.279 Michael Darden: Yeah, so the band comes in, these bands come in and change that that market in some interesting ways, potentially. Can you speak to that at all? 139 00:22:21.390 --> 00:22:33.579 Andreas Stoller: Yes, what I know about it is that it's a quite intransparent market, like, it's… at least for me, it's difficult to find information, I mean, about the industry. 140 00:22:35.140 --> 00:22:45.450 Andreas Stoller: overall. And with the billboards, there is an association that That gives some. 141 00:22:47.090 --> 00:22:57.330 Andreas Stoller: very aggregated, like, general information about the… about billboards in Switzerland. However, I don't have the funds to… 142 00:22:57.520 --> 00:23:02.880 Andreas Stoller: ask for this information, because it's quite costly, actually. So… 143 00:23:03.250 --> 00:23:11.309 Michael Darden: I guess I guess another way of saying the question, like, you know, the the billboards that are specific to tobacco products go away. 144 00:23:11.730 --> 00:23:13.820 Michael Darden: Ryan Miller, What takes their place? 145 00:23:13.950 --> 00:23:20.720 Andreas Stoller: Yeah, I mean, I think it's a valid question, and I looked into that, but it's… 146 00:23:20.830 --> 00:23:27.589 Andreas Stoller: For me, it's not feasible, it was not feasible to get a hold of this information. 147 00:23:28.800 --> 00:23:36.939 Andreas Stoller: One might speculate, but… Yeah, here I don't… like, at the moment, I don't feel… but… 148 00:23:37.390 --> 00:23:50.400 Andreas Stoller: what could be concerning. For example, one might think that the alcoholic beverages would be advertised instead, like I thought about such 149 00:23:50.700 --> 00:24:00.340 Andreas Stoller: potential limitations. And this could drive effects, I mean, this is, like, if I would think about the problematic case, then it would be that 150 00:24:00.810 --> 00:24:05.179 Andreas Stoller: However, I… Yeah, sadly, I cannot, like… 151 00:24:06.510 --> 00:24:11.889 Andreas Stoller: advocate in favor or against it, because I don't have the information. 152 00:24:12.650 --> 00:24:20.459 Michael Darden: Okay, yeah, fair enough. Why don't we move on to the rest of the presentation and we'll hear about the strategy and the results. 153 00:24:21.560 --> 00:24:22.480 Andreas Stoller: Sure. 154 00:24:30.730 --> 00:24:44.130 Andreas Stoller: Okay, now I should be sharing my screen again, and I can just continue with the estimation part. So, for estimation, I use a difference-in-differences approach. 155 00:24:44.730 --> 00:24:46.220 Andreas Stoller: Here we, 156 00:24:48.770 --> 00:25:04.550 Andreas Stoller: with an event study design. So this allows me to estimate dynamic effects on the smoking rate up to 5 years after the billboard ban introduction. I will also look at heterogeneous effects by gender and age, and 157 00:25:04.660 --> 00:25:13.489 Andreas Stoller: An important part that I haven't mentioned yet, but I will not only control for gender and age, but also for other 158 00:25:13.720 --> 00:25:22.369 Andreas Stoller: tobacco prevention policies that happened at the cantonal level. So, there were these sales bans for minors that were introduced. 159 00:25:22.530 --> 00:25:39.500 Andreas Stoller: And also, smoking, bans in public spaces that were introduced at similar times. It's not completely overlapping with the billboard bans, but, there is certainly some correlation, present, as… 160 00:25:39.900 --> 00:25:49.569 Andreas Stoller: Like between 2000 and 2010, Cantons decided to introduce a few of these tobacco prevention measures. 161 00:25:50.470 --> 00:25:56.739 Andreas Stoller: The way I will analyze the data is using a staggered difference-in-differences estimator. 162 00:25:57.130 --> 00:26:08.929 Andreas Stoller: this estimator is preferred over the two-way fixed effects estimator because we have a staggered introduction of the treatment. So, some regions 163 00:26:09.230 --> 00:26:17.490 Andreas Stoller: apply, introduce the… policy earlier than others, and with two-way fixed effects, we would 164 00:26:17.810 --> 00:26:32.579 Andreas Stoller: risk to have negative weights introduced. This is a common problem in economics, so that's why we now use this staggered difference-in-differences estimator, like here, from Callaway and Santana. 165 00:26:32.700 --> 00:26:45.309 Andreas Stoller: But there are also others, but this is one of the main estimators one uses for this problem. And, of course, it requires common trends and other, like, standard difference-in-differences. 166 00:26:45.390 --> 00:26:56.390 Andreas Stoller: assumptions. And at the end, in the robustness checks, I will talk more about assumptions and how… what are limitations and what I could check. 167 00:26:58.270 --> 00:27:09.929 Andreas Stoller: I just very briefly want to look at the event study results based on the staggered difference-in-differences approach, because actually, even though the… 168 00:27:10.190 --> 00:27:22.180 Andreas Stoller: The pre-treatment trends look okay. In the event study design, the common trends assumption doesn't hold in this case, so, 169 00:27:22.600 --> 00:27:27.420 Andreas Stoller: And this would be required for the validity of this design. 170 00:27:27.560 --> 00:27:47.369 Andreas Stoller: But before I talk about this in detail, I will just interpret the results. So what we see is we have an event study design with 6 periods after the treatment. So these are the effects up to 5 years after the treatment. 171 00:27:47.680 --> 00:27:55.310 Andreas Stoller: Introduction of these billboards, and… We see here a slight reduction in the… 172 00:27:56.140 --> 00:28:08.470 Andreas Stoller: in the smoking rate due to the billboard bans. The strongest reduction is at, at the third, point estimate, namely about 0.9 percentage points. 173 00:28:08.850 --> 00:28:11.640 Andreas Stoller: On average, though, like, over the… 174 00:28:11.750 --> 00:28:18.119 Andreas Stoller: All the post-treatment periods, the reduction is about 0.4 percentage points. 175 00:28:20.130 --> 00:28:27.280 Andreas Stoller: And the point estimates in blue before the introduction of the Billbrook ban are the… 176 00:28:27.480 --> 00:28:42.069 Andreas Stoller: pre-treatment trends, so they help us to see if the common trend assumption holds. And actually, here, it looks like the common trend assumption holds, maybe with one or two exceptions, but 177 00:28:42.870 --> 00:28:51.450 Andreas Stoller: More strictly speaking, it has to hold in all the treatment groups, so all the regions introducing the 178 00:28:51.670 --> 00:29:01.580 Andreas Stoller: Wilburben at a different point in time. For all these regions, the, the common trends assumption must hold, or yeah, the pre-treatment 179 00:29:01.580 --> 00:29:12.179 Andreas Stoller: trends must be undistinguishable from zero. And if I run that F-test, the common trends assumption doesn't hold. And my… 180 00:29:13.260 --> 00:29:18.000 Andreas Stoller: My interpretation of this is that there is 181 00:29:18.670 --> 00:29:21.370 Andreas Stoller: A lot of interventions going on. 182 00:29:21.550 --> 00:29:26.910 Andreas Stoller: At the regional level that we cannot capture, we cannot control for. 183 00:29:27.020 --> 00:29:40.600 Andreas Stoller: There are leagues… there are organizations, NGOs against cancer, or for the youth, and there are also, like, intervention programs organized by the cantons. 184 00:29:40.610 --> 00:29:52.159 Andreas Stoller: And we tried to get a hold of this information, but sadly, it was not feasible to have a systematic collection of this tobacco prevention going on. 185 00:29:52.230 --> 00:29:59.350 Andreas Stoller: So, my, assumption, or my guess here would be that 186 00:29:59.490 --> 00:30:07.020 Andreas Stoller: this unobserved prevention going on that might be heterogeneous across Switzerland. 187 00:30:07.180 --> 00:30:10.539 Andreas Stoller: leads to common trends not holding. 188 00:30:10.860 --> 00:30:18.729 Andreas Stoller: And this is why I use another estimator which wants to overcome this problem. 189 00:30:18.860 --> 00:30:34.239 Andreas Stoller: Namely, the interactive fixed effects counterfactual estimator. I don't want to go into details with that estimator, but you have to think about it like a synthetic control approach. Actually, you want 190 00:30:34.420 --> 00:30:38.630 Andreas Stoller: To, learn a model that explains the, 191 00:30:39.020 --> 00:30:51.920 Andreas Stoller: the outcome of the treated observations in the post-treatment period if they were not treated. So, you really want to model this counterfactual of the treated post-treatment observations. 192 00:30:51.920 --> 00:31:01.269 Andreas Stoller: And you do that by training a factor model using all control observations and the pre-treatment treated observations. 193 00:31:01.450 --> 00:31:07.810 Andreas Stoller: And in this… Using this data, you want to 194 00:31:08.210 --> 00:31:16.500 Andreas Stoller: Have a model that explains the outcome within this data that should not be affected by treatment. 195 00:31:17.780 --> 00:31:30.720 Andreas Stoller: And the twist here is that you don't only do that, but you allow for latent factors. So you model factors, covariates, that 196 00:31:31.550 --> 00:31:38.520 Andreas Stoller: that are discovered by an algorithm. And Which… 197 00:31:39.210 --> 00:31:49.410 Andreas Stoller: which will find patterns that are the same for all the individuals, however differ at an individual level. And, for, 198 00:31:49.750 --> 00:32:06.140 Andreas Stoller: for this, I want to show you a graphical illustration of what we are actually trying to discover with this latent factor model. So, the assumption is that we have a factor that is the same for all the individuals. 199 00:32:07.430 --> 00:32:09.290 Andreas Stoller: And that changes our time. 200 00:32:10.260 --> 00:32:18.440 Andreas Stoller: However, this factor is more or less present for each of the individuals, also called, like, the loadings. 201 00:32:18.590 --> 00:32:34.310 Andreas Stoller: and this is all hidden. This is all in the in the error for the model without the latent factors, and the this structure is then discovered by by this by this estimator. 202 00:32:35.000 --> 00:32:50.420 Andreas Stoller: I want to… I mean, you cannot really know what it's capturing, but just to give you an intuition about it, we might have a change in smoking sentiments across time. However, this says 203 00:32:50.420 --> 00:33:03.000 Andreas Stoller: attitudes towards smoking, they… they might be, more changing in some… for some individuals than others. Maybe people who read a lot of media, they, 204 00:33:03.790 --> 00:33:09.470 Andreas Stoller: or reliable media, they might be more impacted by that than people who don't. 205 00:33:10.760 --> 00:33:27.809 Andreas Stoller: So, this is the main idea of this estimator, that you discover this model of the outcomes, so that you then can impute the counterfactual outcome of the post-treatment treated observations. 206 00:33:28.540 --> 00:33:39.890 Andreas Stoller: And the treatment effect then is just calculated taking the difference between observed outcome of the treated and the imputed counterfactual outcome. 207 00:33:41.620 --> 00:33:55.380 Andreas Stoller: So, yeah, I know this, method is maybe rather novel in, in economics. However, now Deschez-Martin and Toth-Foy, they… they discuss this estimator in their new, DID book. 208 00:33:55.470 --> 00:34:08.139 Andreas Stoller: that is, I think, online… on a repository for the moment, not yet published. I just recommend you to… to read that book. It's a really great one with many Stata applications. 209 00:34:08.270 --> 00:34:18.389 Andreas Stoller: And yeah, I'm happy to provide one of the first such applications in economics. The method comes with a drawback. 210 00:34:19.150 --> 00:34:21.440 Andreas Stoller: Namely, that we have many 211 00:34:21.560 --> 00:34:32.400 Andreas Stoller: pre-treatment periods that we need to learn this model, and also that this latent factor structure must be, like, correct and not too complex. 212 00:34:32.540 --> 00:34:39.620 Andreas Stoller: I forgot the most important part about this model, namely… 213 00:34:39.940 --> 00:34:46.899 Andreas Stoller: what it tries to capture, and that's why we have these loadings, so the factors, the… 214 00:34:47.179 --> 00:34:51.250 Andreas Stoller: The intensity of the factors, how the individuals experience it. 215 00:34:51.710 --> 00:35:01.409 Andreas Stoller: differs across the individuals, precisely because we want to relax the difference-in-differences assumption, that… the common trends assumption, that 216 00:35:01.900 --> 00:35:15.799 Andreas Stoller: So that now we have common trends that just have to hold at the individual level. We can break it down using this latent factor model. 217 00:35:15.980 --> 00:35:23.539 Andreas Stoller: That's the main reason why we want to, to use this, this new estimator in order to, to capture this, 218 00:35:23.730 --> 00:35:28.999 Andreas Stoller: these common trends at the individual level. 219 00:35:30.040 --> 00:35:38.380 Andreas Stoller: Okay, so I think I talked enough about this method, and I will just jump to the results based on this 220 00:35:38.520 --> 00:35:45.900 Andreas Stoller: this latent factor model. And here we see a very different picture. So we still have the 221 00:35:46.190 --> 00:35:54.419 Andreas Stoller: pre-treatment trends, which are the differences between observed, treated before treatment, and the 222 00:35:54.630 --> 00:35:57.850 Andreas Stoller: counterfactual, based on the latent factor model. 223 00:35:58.330 --> 00:36:02.499 Andreas Stoller: And we see that, Then, these effects are… 224 00:36:03.300 --> 00:36:06.380 Andreas Stoller: not distinguishable from 0 for the most part. 225 00:36:06.940 --> 00:36:15.659 Andreas Stoller: meaning that the latent factor model actually managed to capture quite well the the hidden structures that were around. 226 00:36:16.790 --> 00:36:24.499 Andreas Stoller: The outcome doesn't always look like this. Sometimes the pre-treatment trends, they… 227 00:36:24.940 --> 00:36:26.990 Andreas Stoller: They they aren't 0 at all. 228 00:36:27.240 --> 00:36:34.959 Andreas Stoller: Mostly because one has too little information, so one really needs a lot of pre-treatment strands for applying this technique. 229 00:36:35.600 --> 00:36:50.059 Andreas Stoller: Then, I mean, the most interesting part for this seminar, we see that now we have a reduction in the smoking rate of about 0.9 percentage points. That is immediate, and that stays the same. 230 00:36:50.900 --> 00:36:54.960 Andreas Stoller: This is about the reduction of 3% in the smoking rate. 231 00:36:57.180 --> 00:37:00.899 Andreas Stoller: Yeah, so this is the main result. 232 00:37:01.630 --> 00:37:06.380 Andreas Stoller: I can go over to the heterogeneity analysis, and 233 00:37:06.550 --> 00:37:16.059 Andreas Stoller: here we see that the… and this is the heterogeneity analysis, all the robustness checks are then all based on the staggered DID estimator. 234 00:37:16.200 --> 00:37:18.679 Andreas Stoller: just to disclose this. 235 00:37:18.870 --> 00:37:22.079 Andreas Stoller: Here we, we see that 236 00:37:22.710 --> 00:37:31.879 Andreas Stoller: The effect is mostly driven by women, individuals aged 25 to 44, and above 65. 237 00:37:35.710 --> 00:37:39.689 Andreas Stoller: The heterogeneous effects are a bit puzzling, because 238 00:37:40.320 --> 00:37:56.209 Andreas Stoller: One usually would expect an effect among the youth, because the typical argument for these advertising restrictions is that the youth is persuaded, or would be persuaded, to smoke. 239 00:37:56.410 --> 00:38:03.940 Andreas Stoller: But this is precisely not the subpopulation where we find any effects. 240 00:38:04.360 --> 00:38:08.450 Andreas Stoller: So, how do I explain this? Maybe just… 241 00:38:08.580 --> 00:38:15.199 Andreas Stoller: Other factors are more important for the youth, such as the price or social influences. 242 00:38:16.630 --> 00:38:23.550 Andreas Stoller: Maybe another one, and I think this is a bit unfair to discuss, and I came up with a few other. 243 00:38:23.930 --> 00:38:28.600 Andreas Stoller: thoughts about interpreting this for today's seminar. 244 00:38:28.990 --> 00:38:38.849 Andreas Stoller: My… another reason might be that we have substitution from billboards to other advertising channels, even if this is, 245 00:38:40.160 --> 00:38:48.369 Andreas Stoller: limited because of the expenditure information that that we have. 246 00:38:49.030 --> 00:38:59.800 Andreas Stoller: there might be still some… still such effects going on. I… I don't know if the youth will read newspapers, but, 247 00:38:59.970 --> 00:39:14.010 Andreas Stoller: maybe this is still a possibility. Yet, something that just comes to my mind right now, maybe another factor might be that advertising targeted to the youth is already 248 00:39:14.200 --> 00:39:14.930 Andreas Stoller: And. 249 00:39:15.210 --> 00:39:22.540 Andreas Stoller: banned at the national level, so you are not allowed to have advertising that would be targeted at the youth. 250 00:39:24.060 --> 00:39:29.109 Andreas Stoller: However, yeah, that… there I… I like a bit the… 251 00:39:30.360 --> 00:39:41.520 Andreas Stoller: The information, again, on how does advertising actually… did actually look like back then, because it's very difficult to, get a hold on this information. 252 00:39:41.900 --> 00:39:44.540 Andreas Stoller: So, this would be, 253 00:39:44.660 --> 00:40:00.230 Andreas Stoller: my interpretation of why we don't find an effect there among the youth. And I would be really happy to discuss this point because I think it's very important to know how to interpret now these heterogeneous results. 254 00:40:00.830 --> 00:40:03.439 Andreas Stoller: And to draw the correct conclusions. 255 00:40:04.020 --> 00:40:06.890 Andreas Stoller: And for the second part, 256 00:40:07.320 --> 00:40:15.060 Andreas Stoller: Why do we find these effects among women and especially like individuals aged 25 to 44 and 257 00:40:15.960 --> 00:40:18.650 Andreas Stoller: They aren't, you know… There. 258 00:40:19.730 --> 00:40:24.730 Andreas Stoller: My interpretation would be that, These, these groups are… 259 00:40:25.030 --> 00:40:36.390 Andreas Stoller: likely to consider smoking cessation. I mean, if you are… usually people stop smoking once they are older, so this would be a bit this age group, and also the 260 00:40:36.390 --> 00:40:48.189 Andreas Stoller: are reasons like family planning or pregnancy, which might make this group especially interested in stopping smoking. 261 00:40:49.740 --> 00:40:55.850 Andreas Stoller: And the mechanism there, I mean, and this is just speculation, might be that… 262 00:40:56.190 --> 00:41:15.730 Andreas Stoller: less exposure to tobacco advertising may facilitate the smoking cessation. So we have a population that is more likely to want to stop smoking and this type of policy might help them to actually stop smoking by less exposure to advertising. 263 00:41:15.930 --> 00:41:19.290 Andreas Stoller: Again, here I'm very happy to discuss this point. 264 00:41:20.890 --> 00:41:29.210 Andreas Stoller: So, I see that I have about 5 minutes left, so I want to lose a few words on the robustness checks and limitations. 265 00:41:29.440 --> 00:41:39.269 Andreas Stoller: So, as we are looking at this smoking history of the individuals and calculate back their smoking status. 266 00:41:40.330 --> 00:41:53.960 Andreas Stoller: I provide a robustness check just calculating back the smoking status for five years, 10 and 15 years and results are very stable across all these specifications. 267 00:41:54.820 --> 00:42:06.149 Andreas Stoller: One can also use, instead of the not yet treated, the never treated units as controls using staggered DID, and again, we have similar results. 268 00:42:07.150 --> 00:42:20.189 Andreas Stoller: And using different covariate sets, so none, only individual and only policy covariates and a model where all the covariates are interacted. 269 00:42:20.500 --> 00:42:25.729 Andreas Stoller: The only sensitivity of the treatment effect is 270 00:42:26.330 --> 00:42:29.749 Andreas Stoller: Shown once we add the policy controls. 271 00:42:32.310 --> 00:42:43.720 Andreas Stoller: An important part are the spillover effects, or the possible spillover effects. Switzerland is really small, you can cross the country within a few hours. 272 00:42:43.960 --> 00:42:52.380 Andreas Stoller: People are very mobile across these continents, so if you are in a population that is, 273 00:42:52.380 --> 00:43:06.680 Andreas Stoller: here not considered as exposed to the billboard ban, you might still commute to a city where there is the ban, and the other way around. So, you might have such spillover effects. 274 00:43:06.820 --> 00:43:09.340 Andreas Stoller: and the, 275 00:43:10.800 --> 00:43:20.170 Andreas Stoller: I, would have assumed an underestimation, due to these spillover effects, and now I just don't… just have to assume it, actually. 276 00:43:20.400 --> 00:43:32.150 Andreas Stoller: Like recently, I performed a check for spillovers, adding a dummy variable, which is one if a neighboring Canton introduced the… 277 00:43:32.360 --> 00:43:46.389 Andreas Stoller: billboard ban, and providing this, check that, again, wasn't, yet shown to the discussant. Actually, the treatment effect about doubles. So, 278 00:43:47.330 --> 00:43:54.509 Andreas Stoller: The spillover effects actually really tend to underestimate the treatment effect. 279 00:43:55.220 --> 00:44:00.169 Andreas Stoller: And once we adjust for it, the treatment effect about doubled. 280 00:44:01.880 --> 00:44:18.740 Andreas Stoller: Then, for confounding, I mean, I said already enough about the confounding. Sadly, not all prevention efforts can be tracked, and this is, like, a major limitation. Maybe in future, one can come up with such a systematic 281 00:44:19.070 --> 00:44:21.690 Andreas Stoller: collection of prevention. 282 00:44:21.920 --> 00:44:25.199 Andreas Stoller: So now I want to wrap up. The… 283 00:44:25.860 --> 00:44:31.029 Andreas Stoller: Evidence here suggests that I both staggered the ID and this. 284 00:44:31.150 --> 00:44:38.939 Andreas Stoller: Other latent factor, model show consistent results, pointing both to a reduction in smoking rates. 285 00:44:39.460 --> 00:44:48.829 Andreas Stoller: The reduction is up to 0.9 percentage points, or 3%, as the share of smoking is about 30%. 286 00:44:49.590 --> 00:45:02.650 Andreas Stoller: And this effect is comparable to that one of a 12% increase in, I should have written, tobacco cigarette prices. So, the effect is quite substantial. Like, it's really, 287 00:45:02.980 --> 00:45:06.400 Andreas Stoller: In my point of view, it's a considerable effect. 288 00:45:06.740 --> 00:45:08.959 Andreas Stoller: And we have to discuss them. 289 00:45:09.370 --> 00:45:18.080 Andreas Stoller: effect heterogeneities. So finally, we see that there is support of the market expansion hypothesis. 290 00:45:18.500 --> 00:45:27.789 Andreas Stoller: So, advertising, or let's say advertising bans, they actually can reduce, the total demand for smoking. 291 00:45:28.990 --> 00:45:41.990 Andreas Stoller: And thus, advertising restrictions can be an effective complement to other prevention tools. So, yeah, thank you for your attention, and I'm really looking forward to the discussion and further 292 00:45:42.320 --> 00:45:44.379 Andreas Stoller: questions by the public. 293 00:45:45.820 --> 00:45:52.389 Michael Darden: Thank you so much, Andreas. That was terrific. We will go to our discussants now, Dr. Christian Seitz. 294 00:45:53.960 --> 00:45:58.610 Christian Saenz: Thank you. Really interesting presentation, Andreas. 295 00:45:58.810 --> 00:46:03.339 Christian Saenz: A few comments and then a question. So 296 00:46:03.700 --> 00:46:09.530 Christian Saenz: I think both in the paper and presentation, it might be, helpful to have 297 00:46:09.560 --> 00:46:24.549 Christian Saenz: just an example of a typical tobacco billboard, to better understand what… what, drivers and consumers are exposed to. I know in the U.S, you know, back in the day. 298 00:46:24.890 --> 00:46:39.190 Christian Saenz: a lot of tobacco advertising would have, you know, cowboys, you know, with cigarettes, you know, riding in the sunset. So it'd be interesting to know how the, tobacco marketing varies in this country. 299 00:46:39.190 --> 00:46:47.199 Christian Saenz: that you study and have a better understanding of what consumers are exposed to. 300 00:46:47.270 --> 00:46:52.389 Christian Saenz: I thought you addressed the confounding points. 301 00:46:52.610 --> 00:47:02.360 Christian Saenz: In a thoughtful way. I mean, you, you mentioned that, you know, you can't control for every possible confounding policy. I do… 302 00:47:02.770 --> 00:47:19.660 Christian Saenz: wonder what your thoughts are, though. So it… let's say that in the absence of other tobacco control policies and public health interventions, that are… that you control for, but do you think that… that your estimates 303 00:47:19.950 --> 00:47:26.589 Christian Saenz: Are perhaps an upper bound and that the tobacco, restrictions. 304 00:47:26.780 --> 00:47:41.280 Christian Saenz: that they complement these other, interventions. And so, perhaps in the absence of these other interventions, the, policy that you study would have a smaller, effect on smoking. 305 00:47:41.820 --> 00:47:43.440 Christian Saenz: Do you have any thoughts on that? 306 00:47:45.220 --> 00:47:53.020 Andreas Stoller: Yeah, thank you for the, compliments first, and, the, I mean… 307 00:47:53.220 --> 00:47:58.130 Andreas Stoller: I saw a few pictures from back in the times and the 308 00:47:58.930 --> 00:48:04.399 Andreas Stoller: I mean, I don't know how it is to live in the US, but… 309 00:48:04.650 --> 00:48:08.830 Andreas Stoller: It seemed to me like that the billboards are quite. 310 00:48:09.210 --> 00:48:17.100 Andreas Stoller: similar to what you would expect in the US. I mean, now I don't want to say brand names and so on, but, 311 00:48:17.280 --> 00:48:22.349 Andreas Stoller: The, yeah, the picture of being. 312 00:48:22.630 --> 00:48:39.650 Andreas Stoller: Like, these lifestyle arguments, or these, being, like, back then, a strong man, even though that's a bit now an outdated idea, but this was much put forward there, back then. Like… 313 00:48:41.180 --> 00:48:42.640 Andreas Stoller: the type of… 314 00:48:43.020 --> 00:48:49.609 Andreas Stoller: advertising that I saw from back then in the US, and here it seems to be fairly similar. 315 00:48:50.250 --> 00:48:52.470 Andreas Stoller: I'm… And… 316 00:48:53.610 --> 00:49:08.440 Andreas Stoller: Right, yeah, I should show pictures, however, for, like, copyright reasons, I don't know, like, how much one is allowed to share. It would be nice to have one, but these can be quite big pictures, like billboards at, 317 00:49:08.650 --> 00:49:14.100 Andreas Stoller: at, at the wall of buildings. 318 00:49:14.780 --> 00:49:33.000 Andreas Stoller: It won't be the Times Square, but this can be quite big billboards that are shown, or just, like, smaller ones that you see using the sidewalk. Like, we use quite a lot the sidewalk, so, yeah. 319 00:49:33.060 --> 00:49:45.470 Andreas Stoller: Especially in the cities, so, yeah, that's, that's where you would encounter it. Maybe something that has to be specified here, like, it's the public spaces, so… 320 00:49:46.490 --> 00:49:49.570 Andreas Stoller: even though I… 321 00:49:50.070 --> 00:49:57.510 Andreas Stoller: Yeah, yeah, I would have never, like, seen one, but in commercial centers, in shopping centers, like, you… 322 00:49:58.240 --> 00:50:07.419 Andreas Stoller: you might be still able to put on these billboards. However, effectively, I'm not sure if it was actually done. And the… 323 00:50:07.750 --> 00:50:18.930 Andreas Stoller: the expenditures for billboards would tell me, like, that they didn't try to substitute, such billboards in shopping centers. Then we'll, yeah. 324 00:50:19.250 --> 00:50:32.689 Andreas Stoller: With these, billboards. But yeah, again, I don't know enough about these details. I think this is now clear. It's really difficult to have, like, 325 00:50:33.560 --> 00:50:39.990 Andreas Stoller: an accurate idea about how these billboards looked like in these times. 326 00:50:40.620 --> 00:50:41.440 Andreas Stoller: And. 327 00:50:41.710 --> 00:50:45.810 Andreas Stoller: Then, for the upper bound, I mean, I would rather, 328 00:50:46.290 --> 00:51:00.820 Andreas Stoller: argue for estimating a lower bound. I mean one reason is this of the of the spillover effects which tend to underestimate the effect. 329 00:51:01.020 --> 00:51:04.670 Andreas Stoller: Now, 330 00:51:04.990 --> 00:51:20.339 Andreas Stoller: now I have to think that there were a few reasons why now I sadly cannot remember all of them, but the… there are a few reasons why, I think I rather have an underestimation. I mean, especially the spillover effects, it's a shame that 331 00:51:20.340 --> 00:51:30.760 Andreas Stoller: I haven't yet put it into the working paper, but once I adjust for spillover effects, then the effect doubles, so it becomes… 332 00:51:30.770 --> 00:51:37.600 Andreas Stoller: An almost 2 percentage points reduction. So 333 00:51:38.030 --> 00:51:53.309 Andreas Stoller: If anything, I think that I underestimate the treatment effect. But I see your point. I mean, having these billboard bands plus then other prevention measures. 334 00:51:53.530 --> 00:51:58.689 Andreas Stoller: This might rather overestimate the treatment effect. 335 00:51:59.070 --> 00:52:03.380 Andreas Stoller: Given the… the big treatment effect that I measure, I… 336 00:52:05.450 --> 00:52:09.750 Andreas Stoller: I, I doubt that it would be as much as, 337 00:52:09.870 --> 00:52:14.329 Andreas Stoller: Yeah, as to offset the the whole effect. 338 00:52:14.820 --> 00:52:15.560 Andreas Stoller: Yeah. 339 00:52:17.100 --> 00:52:21.669 Christian Saenz: Okay. And then one other comment was. 340 00:52:21.830 --> 00:52:32.889 Christian Saenz: And I think someone in the chat beat me to this, but do you happen to know if e-cigarettes or other nicotine products 341 00:52:33.060 --> 00:52:40.199 Christian Saenz: were included in this policy that you study or or. 342 00:52:42.130 --> 00:52:54.070 Christian Saenz: Excuse me. Or can you also speak to, overall trends in e-cigarette use? And is it possible that, that this tobacco billboard ban 343 00:52:54.070 --> 00:53:06.619 Christian Saenz: Could have been effective at reducing smoking, in part because of the, increasing availability of, of, substitute products to cigarettes. 344 00:53:07.540 --> 00:53:10.589 Andreas Stoller: And a very good point, and I think it's something that 345 00:53:10.700 --> 00:53:15.599 Andreas Stoller: Yeah, I should make more transparent. In Switzerland. 346 00:53:15.770 --> 00:53:22.230 Andreas Stoller: For a long time, e-cigarettes weren't commercially available, like, until 2018, so… 347 00:53:22.350 --> 00:53:42.120 Andreas Stoller: However, people could import them. So even though it's not commercially available, we see that people already consume it in the year 2018. Data before on e-cigarette consumption isn't really present. 348 00:53:42.700 --> 00:53:54.589 Andreas Stoller: I, and cantonal, like, cantonal legislation on this was only enacted much later, like from 2018 on. 349 00:53:55.520 --> 00:54:00.069 Andreas Stoller: Because, I mean, if it's not commercially available, you cannot regulate it. 350 00:54:02.810 --> 00:54:07.260 Andreas Stoller: Another detail that I wanted to add is… 351 00:54:12.060 --> 00:54:16.539 Andreas Stoller: Oh, sorry. I now forgot my train of thought. 352 00:54:18.740 --> 00:54:26.289 Andreas Stoller: But, yeah, e-cigarettes don't seem to, like, be relevant for this, this specific, application. 353 00:54:27.660 --> 00:54:37.370 Andreas Stoller: And yeah, the general population, that's it. The general population tends to pick up e-cigarette consumption very slowly. 354 00:54:37.370 --> 00:54:52.950 Andreas Stoller: what we see is that rather the youth, like, really the school children, adopt e-cigarette consumption pretty quickly. However, among the general population with newer data, you have maybe consumption rates of a few percent. 355 00:54:53.690 --> 00:54:56.690 Andreas Stoller: Yeah, that was the point that I forgot. 356 00:54:56.690 --> 00:54:57.480 Christian Saenz: Thank you. 357 00:55:00.450 --> 00:55:06.810 Michael Darden: There are several questions in the in the chat. I mean, I'll start by just saying, you know this. 358 00:55:07.080 --> 00:55:24.350 Michael Darden: Safer and Chalupka argument about market expansion, right? You would expect to see that in the youth, right? Because they're the ones who initiate tobacco, usage. So, if I… tell me if I'm wrong, but what I… I'm understanding this comment… this, 359 00:55:24.350 --> 00:55:28.910 Michael Darden: This this mechanism is more like the billboards are 360 00:55:28.910 --> 00:55:36.110 Michael Darden: preventing cessation, and when the billboards go away, more people are quitting. Is that… is that consistent with your results? 361 00:55:38.310 --> 00:55:50.880 Andreas Stoller: I think that's the argument I want to make, even though it's rather unconventional, so that's why I'm open to, like, other suggestions that could explain this. Yeah. 362 00:55:51.220 --> 00:56:03.519 Michael Darden: Yeah, well, I mean, I think that has to be it, because you're not. It's extremely rare to see someone start using tobacco in middle age. It's almost always at young people, and you find nothing for young people. 363 00:56:03.560 --> 00:56:22.249 Michael Darden: The factor model is really interesting to me. And, you know, it's, I think, new for TOPS. And so I want to just clarify that I understand it and maybe clarify for the audience as well. So what you're saying, and so I'm going to try to explain it, and you tell me if I'm right. 364 00:56:22.250 --> 00:56:32.320 Michael Darden: What you're saying is that there is a latent factor, a single dimensional factor that is related to the propensity to smoke. 365 00:56:32.510 --> 00:56:37.859 Michael Darden: It varies across individuals by a factor loading. 366 00:56:38.040 --> 00:56:39.830 Michael Darden: Which you are estimating. 367 00:56:40.090 --> 00:56:55.220 Michael Darden: But it's it's a linear term. Right? So my relationship to you is in some way. So we you and I might be different in our factor loadings in a linear way. Is that is that correct? 368 00:56:57.920 --> 00:57:08.260 Andreas Stoller: There is some, like, it's all, correct, what you said, I mean, exactly, I couldn't have said it better, 369 00:57:10.070 --> 00:57:13.260 Andreas Stoller: There is yet another mechanism there. 370 00:57:13.700 --> 00:57:19.409 Andreas Stoller: The this estimator might also end up using no factor. 371 00:57:19.770 --> 00:57:24.600 Andreas Stoller: Like, no latent factor. There, there is, 372 00:57:25.060 --> 00:57:33.069 Andreas Stoller: There is a procedure there that selects the amount of factors that are included. 373 00:57:33.250 --> 00:57:44.450 Andreas Stoller: there can be none, or there can be multiple. And if they… there are, like, way too many, then it becomes too complex. And there is, like, 374 00:57:44.880 --> 00:57:59.409 Andreas Stoller: an error between the predicted outcome and the the actual outcome that is used as like a target value. And you just use the amount of latent factors that minimizes this like. 375 00:57:59.830 --> 00:58:03.120 Michael Darden: Okay, so that is coming from the data then. 376 00:58:03.120 --> 00:58:05.490 Andreas Stoller: If you have multiple ones, then… 377 00:58:05.670 --> 00:58:11.800 Andreas Stoller: If I'm not mistaken, it could be also nonlinear, but here it's still… Linear, yeah. 378 00:58:11.800 --> 00:58:16.630 Michael Darden: Okay, okay. And then I guess the final thing quickly. 379 00:58:17.020 --> 00:58:29.200 Michael Darden: You know you talked. You talked about migration. I think you also should probably talk about mortality. Right? So in a panel you're observing the people who don't die from cigarettes. 380 00:58:29.690 --> 00:58:30.899 Michael Darden: And we know that. 381 00:58:31.100 --> 00:58:36.690 Michael Darden: Cigarettes are gonna be really… causally related to health. 382 00:58:36.800 --> 00:58:47.129 Michael Darden: And so that might be important. I think I think you could do an exercise, though, where you bound how big the migration and mortality effects would have to be to nullify your results. 383 00:58:48.260 --> 00:58:52.690 Michael Darden: nick Martorano CWQMC, And and that would be 384 00:58:52.930 --> 00:58:57.220 Michael Darden: you know whether or not these things matter. 385 00:58:59.310 --> 00:59:05.170 Andreas Stoller: You mean, like, less smokers will end up in the old SEM, like, in the… 386 00:59:05.330 --> 00:59:05.990 Michael Darden: Yeah, so… 387 00:59:05.990 --> 00:59:08.800 Andreas Stoller: a sample because they just, I mean… 388 00:59:09.210 --> 00:59:23.460 Michael Darden: Yeah, so I mean, a couple, yeah, so on a couple of dimensions, right? So like, you know, smokers are dying selectively, but also, you know, we know, at least in the US, and I don't know if this is true in Switzerland, but there's been a growing urban rural gap in smoking. 389 00:59:23.630 --> 00:59:29.949 Michael Darden: Nathan Miller, that might be differentially exposed to these band to to billboards. 390 00:59:29.950 --> 00:59:43.929 Michael Darden: at least in the context of the US, we think of billboards on highways being more of a rural kind of thing, but there's a growing gap there. So over this long period of time, you might have selective migration. 391 00:59:44.340 --> 00:59:57.440 Michael Darden: That, that could matter, and you could model that. But we, we, we're out of time, so, I'm gonna kick it back to, Dan Yi, who will take us out and, and advertise the Top of the Tops, which is coming up next. 392 00:59:58.290 --> 01:00:15.489 Danyi Li: Yeah, thank you. We are out of time. However, if you still have burning questions or thoughts for Andrea Stoller, today's presenter, you can join us for Top of the Tops and interactive group discussion. To join, please copy the Zoom meeting room URL posted in the chat. 393 01:00:15.490 --> 01:00:33.319 Danyi Li: and switch rooms with us once this event concludes. And I will leave this similar room open for an extra minute after the end to give everyone a chance to copy the URL, which is bit.ly slash topsmeeting, all lowercase. 394 01:00:33.530 --> 01:00:45.229 Danyi Li: Thank you to our presenter, moderator, and discussant. Finally, thank you to the audience of 91 people for your participation, and have a top-notch weekend.