Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. ======================================================================== # Improving Official Employment Statistics: New Insights from Private-sector Data Authors: André Kurmann, Etienne Lalé Discussant: Jason Faberman Video: https://www.youtube.com/watch?v=HSCV6-Vfc4U&t=5960s ## Talk (01:39:20 – 01:58:04) [01:39:29] left off. Andre, >> well, thank you very much. Thanks for the organizers. uh for having our paper on the program. This is a joint work with Etien La who is also here in the audience. Uh so the first session really [01:39:42] teed it up very nicely uh with both this presentation using the CES and then the other presentation on using uh private sector data. I I I I realize shouldn't call it alternatives but that's what I [01:39:57] have on the slide so that shorter but um what's sort of striking in in what was striking in these two presentations is that both of these data sets on the official side and the private sector side they [01:40:11] face many of the same measurement challenges. Not all of them the same but many are similar. And in particular one of the big challenges in both cases is how you do how do you distinguish between business birth and deaths? How do you separate this from sample churn? [01:40:25] And then the other one is how do you ensure representiveness especially if you have changing response rates especially if you want to go to a granular granular geog geography sector level. Now as it was um discussed in the [01:40:38] in the CES presentation the birth death issue in normal times is not that important. [01:40:44] You can forecast uh with different ways the net birth death but it's exactly around turning points when birth and death become important. It's exactly when as we learned in the pandemic policy makers, businesses want to know [01:40:58] most about what is happening with the labor data. Okay. And so in this paper what we are doing is we try to develop new methods to address this birth test challenge first by um matching at the [01:41:10] establishment level uh for a for payroll data uh the data to alternative signals on business activity so that we can identify business birth and death. And then as a second step because these [01:41:24] business signals that we get uh cover a much larger universe of of businesses we're then trying to use uh machine learning techniques to extend coverage meaning to to predict what is employment [01:41:39] beyond what the payroll sample gives us. Now we then apply these methods to one of the small business data payroll um uh sources. This one is from Homebase. It [01:41:51] was mentioned in Ryan's talk. Uh and and that that data set is is is really focused on small businesses in high turnover service sectors. And it's a perfect um sort of proving grounds for our methods because business birth and [01:42:05] death, they occur primarily for very small businesses. And in these service sectors which are undermeasured anyway in the CES and in other in other data sets such as um such as the ADP as well um small businesses are really important [01:42:19] and as I will show you uh in fact it's exactly in these uh high turnover service sectors where revisions to the CES have uh an play an important role. [01:42:31] So just to give you an idea what I'm talking about here, the red line here shows total non-farm employment revisions in the CES first read relative to the QCW [01:42:44] benchmark. So from a real time perspective, this is really what you would like to know. What is the truth when you get the first read? Well, how how off are you? Okay. And these swings can be quite substantial. We've learned about this over the last few years when [01:42:58] when um when there was a lot of discussion about this. Uh what's interesting is that when you look at um high uh high turnover service sector, so this retail uh professional services, leisure, hospitality and other services, [01:43:13] they um they account for most of these revisions as the blue line shows. Okay. [01:43:19] And it's exactly in these services in these service sectors that small businesses are important. And the other thing that's also interesting here and I I'll get back to this is that there's quite a bit of cyclicality. During upturns you get positive revisions exposed. During downturns you get these [01:43:33] uh negative revisions. Okay. And we argue that may have something to do with uh business birth and death. Okay. So to fix ideas, I'm going to come back to the employment estimator that's also at [01:43:47] the core of the CES um uh statistics as well as in in uh Ryan's presentations for the ADP. I'm just going to write it slightly differently here. employment estimated equals last period's employment plus [01:44:02] um in the numerator employment gains in period t and in the denominator employment um I I should say just employment in period t and then the in the denominator employment in period t [01:44:15] minus one. Now you can decompose this into employment changes that come from uh continuously active businesses. [01:44:23] That's the EA. And then in the um in the denominator you have um losses from businesses that are closing in time in period t. So they're no longer there in period t they have closed. And in the numerator businesses that are reopening [01:44:38] or newly opening in period t. So they were not there in t minus one. Okay. [01:44:42] Okay. And so the challenge as was discussed is not all entries that you see in your in your frame in your data frame are openings and not all exits are closings. And how you try to deal with this is actually quite important, right? [01:44:56] Right. And so in in both the CES and the ADP FRB um the the the core is they just look at uh continuously active businesses and then they do a exposed adjustment with econometric methods um [01:45:09] obviously in very sophisticated forms but sort of a a top-down adjustment for net birth and death. Okay. Now other um uh other research that sort of came up during COVID what they do is they are a bit more selective. They say, "Oh, maybe [01:45:24] I'm going to disregard all of the new entries. I'm just going to set them all to zero essentially, or so the EO all all zero, or I'm not going to even look at closings. I'm only going to look at the continuing um uh businesses." As we [01:45:38] showed in a in a paper that got published last year based on the homebased data that I will that I will use, you get very different estimates on what you do. So the green line is our estimate that will implement the methods [01:45:52] that I'm about to describe. The brown line would be if you only looked at continuing businesses and you would clearly not get the extent of the COVID um decline and recovery uh that happened according to our estimates. And then if [01:46:05] you um only included uh exits the red line, you would underestimate the further out you go because well you would miss all the uh you would miss miss all the continuing businesses that [01:46:18] are actually closings sorry that are not closings that but they're continuing and the orange line you would overestimates because not all the entries are new openings which in Homebase's case it's a growing company you're going to get a [01:46:32] huge overestimate. Okay, so it's really important what you do uh in in this method. All right, so what do we do? We follow our earlier work and we're trying a bottom up approach where we um try to [01:46:46] estimate the probability that you're an opening conditional on entry here and we are trying to estimate the probability that you're a closing condition on exiting the business frame. [01:46:58] Okay. Uh I'll show you in a minute how we do this. We we we try to harness here uh independent info on business activity to make these estimates and then we once we have these estimates we're going to [01:47:11] impose adjustment factors data so that the implied business birth and death rates correspond to what we see in the administrative population data from the BED and BDS. Um of course we only know [01:47:25] that with a lag. So these uh T's here you should think of them as they are lagging. Okay. So when you know what the BEDS is, you're benchmarking and then you're using those benchmark values [01:47:36] going forward. Okay. All right. Now in our earlier paper, we kind of cobbled together independent business information from Facebook postings that individual businesses had and Google [01:47:51] places that gives us essentially whether a business is closed or not. Okay. And that worked quite fine. But there's some fundamental issues. you need several weeks of data after the fact in order to actually make these estimates. So [01:48:05] there's a accuracy timeless timeliness trade-off and you could only really do this for physical store locations uh where this data is obviously available. [01:48:13] So in this paper what we do is we try to use other types of data. So first from home base we now get a self-reported company creation data uh when when the when or dates I should say when the company was created and reasons for [01:48:27] exiting the homebased uh data frame and on an aside we are matching at the establishment level the homebased data to transaction records on spending in those businesses and uh also as an aside [01:48:41] on visits that come from geoloccate geoloccated cell phone pings to these businesses. [01:48:47] to to uh to get independent activity data. Okay. Now, this combination of data then allows us to do a bunch of consistency checks and also uh use actually machine learning techniques to try to estimate these probabilities. Uh I won't go into the details here, but [01:49:02] it's quite interesting. The more data you have, the more you can actually cross-ch checkck whether your estimators are good ones. All right. [01:49:10] Again, I talked about Homebase. What is Homebase? Uh it's a scheduling and time clock software provider used by a lot of businesses here. I say a 100 thousand because these are the ones that we can match that have high quality data. Um so [01:49:24] it's quite a big sample frame. It's it's uh essentially uh an LEHD sample frame for those who are in the know except that you can't follow the workers but you know everything about the firm which worker is with which firm for how long. [01:49:38] And we see um each day we see the hours that they are logging. So we really know whether someone is is working or not. [01:49:46] Okay. It's an excellent test case to assess the performance of our methods because it's mostly covering small businesses. So less than 50 workers in retail, professional business services, lesion hospitality, other services and [01:49:59] it has large sample turnover. So our methods if they work there, we have hope that they might work other in other parts too. Okay. And then as I said we match by name address to independent data about customer spending that comes [01:50:13] from safe graph. So this is from personal finance apps that safegraph triangulates to uh uh particular businesses and then visits from geoloccated cell phone pings from adban and I'm going to talk about this a bit [01:50:26] more later on. Okay. So we use this to create these estimates of uh o u um of whether a business is opening when it enters and whether a business is closing. Uh the sampling or the weights [01:50:41] in our estimators are taken from the QCW very similar to the ADP uh data and we create from a jumping off point which is the QCW. We create weekly estimates over the year to the next QCW that becomes [01:50:56] then available uh later on. So the QCW here is for uh small business uh or yeah small establishments in uh these particular sectors. Okay. So I'm showing here retail trade and leisure and [01:51:07] hospitality estimates. You can see that the red dots that's the QCW employment generally is very close to what our estimates are. So there is some revision or some benchmarking in the language of [01:51:22] the CES that you would do but it's quite surprising how how close we get. So we get these revisions are not very big. [01:51:29] Okay. Generally all right and this is for sort of normal times. In our previous paper we showed that during COVID it actually also did a very good job. All right. Now from this data you can do many other statistics. You could look at hours worked which is something [01:51:44] that is quite undermeasured in official statistics. We have very good hours data uh and it's quite interesting. We have very good wage data. Um we can also look at job creation, job destructions, gross flows in uh into and out of these [01:51:58] businesses. But what I want to show you here is one particular um statistics that we find interesting because we talk about open and close. On the left I show you cumulative openings since early February or midFebruary which is our [01:52:11] jumping off points and on the right cumulative closings. And the green line you can see for 2026 is below even the cumulative openings during the pandemic year 2020 which is sort of interesting [01:52:25] uh saying that small business openings in these service sectors seem to seem to be trending lower uh according to these estimates and for the uh cumulative openings here we are sort of uh on the higher end but not not that much off [01:52:38] from 2024. By the way, this beige line here, that's what's happening during COVID, right? There's a lot of temporary closings and then they reopen and so the cumulative closings go back down. All right, [01:52:51] so the second part of our paper as I mentioned is we're trying to use data um from these uh sources on that give us [01:53:03] activity signals. So visits to these businesses or customer spending to try to predict for a much larger universe of establishments what employment might be. [01:53:15] And so the idea here is that well homebase is not a very big data set and even other data sets right they may be bigger but maybe we lose access to them at some point. The holdup problem that Ryan talked about is there any hope that we could sort of hedge here by having [01:53:28] such a a predictor um and and still produce reliable estimates. Um the idea here is is it's just a machine learning exercise. So I'm going to train on the match data where I have employment and I'm doing this with uh you know kfold [01:53:43] cross validation for for those of you who are doing um um who are doing machine learning to avoid overfitting and then I will uh try to create an extended employment estimates which includes both the labeled and the [01:53:57] non-labeled employment estimates here uh from both the payroll sample and then only the business activity sample and then I can validate did afterwards. Of course, there's a key assumption here that what I'm seeing in the match data [01:54:11] set is similar to what will be in the much larger data set. And that's obviously a big assumption. [01:54:17] Not clear whether that's true. All right. Okay. Now, I will not show you any results. Why? [01:54:24] Because so far, as I write here, this has been a pretty sobering experience. [01:54:29] And I'm going to be very honest about this. Um, we came in with high hopes and so far it's been humbling. All right. [01:54:36] But I think there's hope too. And the issue is it turns out that both Advan and Safegraph these data sources they are not they are not very good. They have unstable sampling frames. Okay. So [01:54:50] for instance with Advan they just you know redef redefine the number of what they call points of interest establishments every once in a while and then they reattribute visits or cell phone pings to those things in in a fashion that we can't actually know how [01:55:05] they do it. They don't want to tell us. It's kind of untransparent. On the safe graph side, you know, the data comes from these personal finance apps that some people use and so it doesn't really give you a full account of what's [01:55:18] happening inside a business and who uses these apps also changes over time. So, lots of challenges. We are currently working or we just got new access to data that is actually uh um for for the [01:55:33] for the actual card swipe that happens where we see all the transactions where we have much more hope that we may get good signals out of it. So stay tuned. [01:55:42] But the point I want to make here is that you know the stability and quality of this independent data is a key requirement when you want to do this type of coverage extension. And for people who have used safe graph or Advan and there's many of us in the room [01:55:56] including us it's a bit of a cautionary tale. You got to be really careful what kind of data you use before you make uh big statements of of of you know I had this causal effect or that causal effect or no effect at all. Okay. All right. So [01:56:09] with this I'm almost done. Um so when we first talked with the conference organizers we were kind of asked to you know we really like to see big bold [01:56:23] ideas big bold new ideas um maybe this is I don't know if it's big and bold uh but we try to take a different approach to this business birth death model uh issue that is is is is nagging [01:56:37] essentially everyone okay and and it's quite successful uh We we are we are quite heartened by by what we've come up with. Of course, for a small data set, it would be very nice to do it for a bigger data set. Um for the other part, [01:56:50] these the the the sample coverage extension that's that's a bigger haul and maybe we will not be successful, but I think it's worth trying for all the reasons that were discussed by John uh by David Bill Cox also when you lose access, you know, is there maybe something else by Ryan? Maybe there's [01:57:04] something else, right? Um the other thing I will say here is that this is really challenging what we do not because of any you know coding or AI does that anyway or whatever else the [01:57:18] really challenging part is to convince these companies not only to give you their data but to give you data on the actual businesses the names and addresses so you can do the matching okay and I have thought long and hard [01:57:32] like how what what what's the best mechanism it's kind of a catch 22 both businesses want to work with you or both of these providers want to work with you but they don't want to share that address data. Um so maybe there's some role for a clearing house. I don't know what the clearing house could be [01:57:46] but maybe there's some good ideas in this audience of how one could set up a good mechanism where everyone would be comfortable sharing the data. So with this thank you very much uh and I look forward to uh the other presentations and our discussion. Thank you. ## Discussion — Jason Faberman (02:43:45 – 02:54:29) *Shared discussion for Session 2: Utilizing Payroll Data; the discussant covers every paper in the session.* [02:43:45] All right. Thank you and uh thank you for the opportunity to uh discuss these papers. Uh I will go through all three of them in as much detail as possible in 10 minutes. [02:43:56] All right. So as you might guess, we're going to talk about more kind of the highle's comments and high level thoughts about these three papers. [02:44:02] Luckily, they all share uh some common themes in terms of what they're trying to do and for me at least my interpretation uh of the three also in in kind of the takeaways uh to varying degrees. Okay. So the key theme here or [02:44:16] key theme here is these papers are focused on using new data to improve on at least on the employer side uh labor market statistics in one way or another. [02:44:24] Uh the Kerman LA paper is looking at private sector business data and how that combined with a specific methodology that focuses on on getting the openings and closings um estimated correctly. uh how that can improve our [02:44:39] our data and and do so also in kind of a a fairly important niche among smaller high turnover businesses which our standard business surveys um tend to either not capture well enough or mismeasure in one way or another. Uh the [02:44:52] Belladall paper is looking at how much we can push um how much we can push private payroll data uh to get much better regional statistics on employment and earnings at uh at a much higher uh geographic detail and also a much higher [02:45:07] frequency. And then the most recent paper with UK data is looking at how we can use not just private sector data but internal administrative data uh to improve what what the UK and and and statistical agencies more broadly are getting from their surveys on the [02:45:22] employer side. Okay. So, I'm going to go through each paper um and contribution and kind of the the main takeaways I I hope we all take away from each one in terms of what they're trying to do and and what the successes are and then I'm going to have [02:45:35] some comments on what the limitations that come about from each one. Okay. So first with the Kerman LA paper um like I said this is using private sector data from three ser three sources the homebased data in addition to Advan and Safecraft to to in partly do a data [02:45:50] validation exercise and in some sense see how well we can capture um movements of employment and earnings um but also to apply this methodology Andre was talking about at getting a better estimate of of openings and closings and [02:46:04] you know it's important to understand as Andre pointed out and I think a lot of people that work with these data know the the private sector data suffer from this problem of being an opportunity sample convenience sample. In other words, um firms are entering and and exiting their their surveys uh or their [02:46:19] data for for no other reason than business related reasons, not being a true opening or closing. The real uh the real contribution here from this paper is putting together a methodology that that better identifies true openings and closings from these data. And andre [02:46:34] didn't show you all the all the results. um he said is a pretty sobering exercise. It they don't do that badly in in some sense. Is it is it a perfect match? Is kind of does it meet the bar for statistical samples? No. But the the the methodology certainly makes some [02:46:47] some progress here. It performs relatively well. Um and and as Andre mentioned there there's this drop off that the data seem to be showing in terms of uh in terms of the creation of new businesses over the last year or so. [02:46:59] Um that hasn't showed up in the in the published uh government statistics, at least not yet. So it' be interesting to see how that looks like postbenchmarking um a little less than a year from now. [02:47:10] The regional employment statistics are are taking data from a from a private payroll survey uh like I said to get these more detailed estimates at the regional and um regional regional level and do it at a weekly frequency. So very high frequency data. That's one of the [02:47:25] benefits they have from from the payroll surveys. You people get paid every week. [02:47:28] So they have they have these uh they have the higher frequency data and potentially have it at a higher level here. um they they they show that um I'll argue how how well it it tracks but the national level um we get similar [02:47:42] estimates in in terms of employment um and and the real uh the way in which the authors are trying to show here the real power of these data is to look at these natural disaster events where where published data statistics or or surveys [02:47:56] aren't necessarily going to get the magnitude of the the true impact of a natural disaster like an earthquake, hurricane, flood and so on. [02:48:05] Okay. And then finally, the most recent paper presented um John and Johannes showed that that that essentially the UK equivalent of the United uh United States LEHD program um can be used to generate labor force statistics on [02:48:18] employment and earnings. Um and and really, you know, if if you've worked with the LHD or worked with any any of the administrative data like the QCW, maybe you found yourself a little bit jealous of of how the data is collected [02:48:31] there. Um, and you should because it's kind of ridiculous how much nicer they have it relative to us. Um, the the the the tax records are essentially a real time ser administrative set data set of [02:48:45] uh of wage records um continuously updated and updated at the national level. So in other words, unlike in the US where the federal government has to essentially herd 51 sheep of labor market uh labor market uh departments at [02:48:59] at the state level to get all the data. um and not only wait for all of it to come in but then process all those data which are collected 51 different ways. [02:49:07] Uh in the UK it's all done at the national level and it's all um all done with a uniform um reporting um reporting paradigm. So they they don't have to worry about any uh any of those issues and as a result they're able to produce [02:49:21] um not only consistent estimates that match up with the survey data very well. [02:49:25] Um but they're also to do it a much timelier way than we would be able to do with either the QCW or LHD here in the US. [02:49:33] Okay. So each of these papers are showing varying degrees of promise. Um all three have limitations one way or the the other. So I'll talk about those for a couple minutes now as I finish up. [02:49:43] So with the criminal LA paper, the usual caveats about using private data sources apply. Um though this methodology really goes to the heart of one of the bigger issues we worry about um with these private sector data sources is that that's that they're a convenience [02:49:58] sample, not necessarily a probability sample. Um and that's that's really kind of the one thing I think to take away here. um they they focus on this this niche of of smaller businesses within particular industries, these higher turnover businesses. But these are also [02:50:11] particular, you know, precisely the ones that we that we tend to care about for measuring um or updating birth death models and and and and the like. It's it's the small business that give us the issues. So there's some promise here. um [02:50:26] you know there's some data measure measurement issues I won't get into but there's there there are the you know there there's real promise here in terms of the methodology if not the particular data set itself now the the Belladol paper with uh the [02:50:39] payroll survey data um I think it's a great exercise I think it's a very useful exercise but for me at least when you dig a bit deeper into the data I'm wondering whether it's up to the task of these weekly regional uh estimates that [02:50:52] that the authors are after um for on we saw in the national level data there in terms of matching up to QCEW um generally it looks similar but in terms of the quantitative differences um for for large period it's on the order of several percentage points in in terms of [02:51:07] the level of of of employment over certain periods and and Kyle didn't show the the numbers but in the paper they also look at these uh correlations over time across states of how well the the series match up um QCW versus the [02:51:21] payroll survey and and the correlations aren't that strong across a large swath of states and even some states negative, right? So, so the the the matching of the data at the at the regional level isn't all that great. Um the other issue [02:51:35] is that the the the nature of the data or or at least the the methodology they use, they focus on continuing establishments and they do that to to get consistency of the growth rates which you need for for consistent growth rates. One of the issues though is on on [02:51:50] the plus side, these data allow them to, as Kyle showed, allows them to evaluate major natural disasters or other major shocks to a local economy in ways that the survey data do not. Um, but at the [02:52:03] same time, they may be missing things. So, so one of the things that came out uh at least when you compare the the payroll survey estimates uh of the two counties following the hurricane uh the payroll survey estimates compared to the QCW is they show a much bigger drop on [02:52:17] impact, but they also show a much stronger recovery. My read of that is that's probably because we're missing all of the exits that may have happened during that period by focusing on the continuous establishment. So unless there's a way to to for example apply [02:52:30] the methodology from the first paper to the payroll survey data, we may be limited in what we can say even at a local level um from from these payroll survey data. And then finally, the the the Bryson and all paper. You know, like [02:52:44] I said, it's hard to not be a little jealous of of the UK uh matched employer employee data because it really does seem they're they're able to do quite a bit um given the nature of how the data is collected um and how well it matches to the surveys. in that, you know, of [02:52:58] all these surveys I've seen across several of these conferences now, this is probably the most promising in terms of something that could either complement or if not completely replace an existing um existing major uh labor [02:53:10] statistics survey. Um you know, obvious the the obvious kind of cheap shot caveat is it's not on the household side. We know in the UK their their household survey has some issues. You're not going to get any information about the non-employed um from this survey. Um [02:53:25] but on the plus side, I think there's more information they can get um more than just the the the uh employment and earnings. Um particularly they can follow in the steps of the Qi uh QWI estimates and and produce some estimates on on worker turnover and the like. Um [02:53:39] just as as we have in the US um because you know the data is given how good the data is for the the basic accurate statistics, it's likely be just as good for those statistics as well. So to summarize, these are three really nice papers that tackle, you know, the the [02:53:54] general theme of how much can we use the these external data sources to improve our at least employer side labor market statistics. Um they all do fairly well and provide provide some useful evidence and useful information and and in some cases useful methodologies for tackling [02:54:08] that. Um but just as you know a lot of other papers in in this um this line of literature has found there's some implementations as well. Thank you. ## Q&A (02:54:29 – 03:10:19) *Shared floor discussion for Session 2: Utilizing Payroll Data.* [02:54:29] So we have we have some time now for uh questions from the audience questions or comments on any of these three papers. [02:54:40] >> Okay. So great great papers. I'd like to ask a question from the UK at least to my eye going all the way down to the bottom of your paper and say why not just do the [02:54:53] full this is very convincing that that you can get these data this way and then if there data which are missing uh there are various ways that these that get get other data that will fill the [02:55:08] gaps and there are basically two routes. one would be enhancing the the wage equity that the data going to the uh to the Treasury and that's that's what they did in Australia. Uh and I hast [02:55:21] if you're going to ask askers for more responses like parttime versus full time or other things that tax administration you have to say quit quote pro we're not going to ask you on on survey and if you [02:55:35] need more from the household side you could you could other say what we say picking the getting the getting the household data say demographics from across the census [02:55:47] data. So I would this is great and it seems like an infrastructure to sort of that could be enhanced to fill fill in a lot of the gaps fairly in a fairly straightforward way. [02:56:00] So may I John or do you want to res respond to Matthew's proposal for a revolution? [02:56:11] >> Thank you. Yeah. So um we absolutely recognize the opportunities there but we also recognize the the responsibility that we have to um not uh upturn the apple cart. I don't know this is a [02:56:25] phrase we have in the UK. So um I think the wholesale replacement is challenging because these as as I briefly mentioned the surveys do provide additional information which is very hard to get [02:56:37] from uh the firms. Now Matthew's response to that was well why not get the tax office just to ask more of the firms and I think you know the pressure for that will will come but the tax [02:56:51] office's job is to administer tax not to generate statistics uh it like any organization has resource constraints and it has to fulfill its main mission. [02:57:04] So that's a challenging conversation to have. Um it's a reasonable ambition I think but it's it's um very difficult in practice. Uh there have been conversations about collecting more detailed hours data, occupation and so [02:57:18] on but um they they usually found on the practicality. Uh there are opportunities however to link the data that we have to other sources. So we can potentially [02:57:31] link to information on a benefit receipt which will allow us to identify when somebody disappears from the employee records is that because they're inactive or unemployed and so on. So that's a [02:57:44] potential for the future. And one of the good advantages of the data that we have is it has uh the ONS has developed some um indexes which allow us to link across [02:57:58] from our data set to quite a wide range of other administrative data set. So I think that will come uh in time. And just to point to Jason's uh prompt about [02:58:10] generating QWI type indicators, job to job flows, that's in our near-term horizon. So we're expecting to publish something out of that in the next six months and you'll see that soon. [02:58:23] >> Great. Thank you. Um >> did you want to add? [02:58:28] >> So I I have uh Bruce and and David. So Bruce, since the mic's over by you. [02:58:34] >> Um, this is actually that same question. So if you link to um surveys, you can possibly take a lot of [02:58:48] the burden away from surveys by providing answers to some of the questions that were previously obtained from surveys and make it easier on the [02:59:02] So that seems like another route that you want to go with this wonderful that >> Yeah. [02:59:17] >> So it seems we lost something in 1776 when we separated. [02:59:24] I admit to having a severe case of Jason Paperman's uh uh what was the word you used? Uh envy. [02:59:33] >> Yeah. >> Um what So what I'd like to hear from our side of the pond actually is what would it take to create what you've built in the United States? [02:59:47] Can somebody talk me out of the delusional notion that we could actually overcome 51 different collection agencies and so forth and so on? So what would be what what would be involved in [02:59:58] getting what you have? >> I think why don't we why don't we let John Haltwanger try to answer that question. [03:00:13] >> It's called the Internal Revenue Service. I think what we do I I I think the by the way we have tremendous data in that this idea of this reporting in real time at such a high frequency we're not going to get there immediately but it is [03:00:27] true payroll tax reports come in on on a on a flow basis and uh takes a very long time for the census bureau to process them to build all of its products and so on but they are coming [03:00:40] through maybe put put on the spot it may that we can improve with working with IRS the processing of that because because the data is actually much more timely in principle uh to be able to start doing things like [03:00:55] this. >> Um Julie >> and I I should have said if people would just introduce themselves when they get the mic. [03:01:09] >> Hi Julie Hunt with Bureau Statistics. Um, so the US, yes, we have two forms. [03:01:15] Obviously, one's the IRS. Um, but any of you who are from around the RS would not like to share. Uh, and once things go to Census Bureau, they never leave, which is great for census, but what the rest of us who want to work with statistics [03:01:28] isn't so great. The other one is through where we're getting the QCW through our state partners. Um, and we've been working a lot with our partners at the Employment Training Administration, having serious conversations with them. [03:01:41] that I was going to talk about a little bit tomorrow too. Um, but there is a movement on that side where a lot of um the the lack of utility or or could be utilized a lot [03:01:54] more is the lack of consistency and a lot of states are adding the enhancement to the wage records and there has been a formalized um methodology or definitions. Uh so they're having serious conversations at the national [03:02:06] side uh that I hope that much sooner than I ever see changing in the tax records. I'm I'm more hopeful on that side. So um yeah, which which mean which [03:02:20] is promising for everyone improving the whole statistical system because historically uh the states share with BLS and then BLS can then share uh with our partners and then we can share back with the states. certain improves the whole infrastructure uh which lifts all [03:02:35] of us up. >> I'm going to add one thing Andrew Foot from Census ID team. Um, one thing that we do have [03:02:47] at um, inside of the US, but the statistical agencies currently do not have access to is the national director of new hires, which is we see flow basics of every hire, like as soon as they fill out an IN9, it goes to the [03:03:01] national diretory of new hires and there has to be legislative changes to make it insurable to census and BLS that that would provide real time statistics on hires. provide very timely UI claims uh [03:03:16] information at the individual level and the UI wages the lag will be less than what we get from the states currently it wouldn't solve the GCW problem but there exists a lot of the information there but it requires a legislative change [03:03:30] because of where it lives within the executive bridge >> yes amen and >> so Eric or BS commissioner um and so you'll be [03:03:47] hearing a little bit more tomorrow about some of the steps that could help close all of these different efforts and uh the two I'm going to be talking about is [03:04:00] an effort to create standards for employers to use at the very And that opens up a number of opportunities for ways to collect the [03:04:12] data, gather them. We talked about that. So enhancing, standardizing the technology of the data and then also what are the changes that have to be made to the statistical system so that they can actually use these data to the [03:04:26] best effect. Just bit of foreshadow. [03:04:33] Any other quick comments before we take a break? [03:04:40] Okay. >> What? [03:04:44] So I thank you. I I I as Erica noted, we'll have a chance to pursue some of these questions a bit more tomorrow. Um why don't we stop here and thank our presenters again. [03:05:00] Uh we we will resume in about 25 minutes at 4:30. [03:06:09] That's kind of I [03:06:22] Get back. I can tell you [03:07:17] right Nice. [03:07:55] recognizer. It might be possible. [03:08:53] like our what we get from my head down. [03:09:23] That's actually angel. [03:09:44] There's more than one. I don't even know.