{ "meta": { "video_id": "AkzO-V4cEhA", "source": "transcript-api manual/en", "target_language": "none" }, "segments": [ { "start": 6.488, "end": 23.668, "text": "So welcome to the Global Network webinars. My name is Alexander Loshki and I'm moderating this webinar today together with my colleague Panya Yu. We have with us today Hussein Hassanim and Steve Mcfeely," }, { "start": 23.668, "end": 33.888, "text": "who will talk about how to support the development of composite indicators using the Composite Index Builder tool for R." }, { "start": 36.088, "end": 50.728, "text": "The webinar today focuses on the challenges and opportunity of building composite indicators that support clear evidence-based decision-making, and our speakers will begin with an introduction to composite indices." }, { "start": 51.128, "end": 71.568, "text": "What they are, why they matter for policy and official statistics, and why their construction is inherently complex. The this complexity is exactly why our colleagues developed a new solution, the Composite Index Builder and open source R package." }, { "start": 73.128, "end": 91.208, "text": "The speakers will then walk us through the features and architecture of the tool and will demonstrate the tool in action using a real world data set showing how the tool transforms raw multi indicator data into interpretable." }, { "start": 91.568, "end": 111.208, "text": "Composite scores and interactive geospatial insights, helping users identify patterns, hotspots, and progress in a way that is both technically robust and immediately actionable. And let me introduce now our speakers. So we." }, { "start": 111.208, "end": 126.455, "text": "We have with us Hosin, who has more than 20 years of experience in data science, official, applied and theoretical statistics. He has worked with many international and intergovernmental organizations and has" }, { "start": 126.455, "end": 144.928, "text": "served on the editorial boards of several journals. Including the Statistical Journal of the International Association for Official Statistics, Hoisin has also authored more than 200 peer-reviewed journal articles and seven books." }, { "start": 146.128, "end": 166.994, "text": "Steve is the chief statistician of the OECD and director of the Statistics and Data Directorate. He's also adjunct professor at the University College of Cork in Ireland. And prior to joining the OECD, Steve served as the Director of Data and" }, { "start": 166.994, "end": 184.655, "text": "Analytics at the World Health Organization and as Director of Statistics and Information at the UN Trade and Development, or known also as UNTAP. Before joining the UN, he was the Deputy Director General at the" }, { "start": 184.655, "end": 204.077, "text": "Central Statistics Office in Ireland. And before I hand over to our speakers, just some practical remarks as always. You can write your comments and questions into the meeting chat at any time, and then during the Q& A session we would encourage you to ask" }, { "start": 204.077, "end": 220.97, "text": "the question yourself, but if you're a little bit shy, we can also do it for you. As always, the Global Network webinar is being recorded and will be made available on the new Global Network of Data Officers" }, { "start": 220.97, "end": 242.238, "text": "and Statisticians, which has transitioned to LinkedIn groups. I will share the link in a second, so if you haven't joined the new global network yet, we invite you to follow that link in the chat and join the group on LinkedIn. And we also invite you to sign up for our" }, { "start": 242.238, "end": 263.008, "text": "newsletter. Also, we will put that link also in the chat and without further ado, now over to you. Steve, the floor is all yours. Thank you. Thanks, Alex, and thanks to you and Sopanu for the invitation. It's great to be back on the." }, { "start": 263.768, "end": 280.288, "text": "On one of your webinars. So good morning, good evening, good afternoon to everybody and thanks for joining us. So as Alex said, my name is Steve Mcfeely. I'm the chief statistician at the OECD and today we're going to be talking about composite indexes." }, { "start": 280.768, "end": 296.392, "text": "The OECD publishes a handbook on constructing composite indicators and indices, and this methodological user guide was last published in 2008. In fact, we're discussing whether we need to update it at the moment. But in that handbook," }, { "start": 296.392, "end": 310.068, "text": "it defines a composite indicator as an indicator that is. formed when individual indicators are compiled into a single index on the basis of an underlying model of the multi-dimensional concept that's being" }, { "start": 310.068, "end": 323.395, "text": "measured. So that's a bit of a mouthful, so we'll come back to that later on. Now, as many of you probably know, statisticians have a bit of a love hate relationship with composite indices. People really," }, { "start": 323.395, "end": 337.008, "text": "it seems to be like Marmite. People either love them or hate them. But whether you love them or hate them, the reality is they get used a lot and some. Some of the better known examples are the UNDP Human Development Index." }, { "start": 338.088, "end": 352.791, "text": "The UN Multidimensional Vulnerability Index, UNCTAD's Productive Capacity Index, their Inclusive Growth Index. Here at the OECD we have a Better Life Index, which I think is pretty well known. We also have the Social Institutions and" }, { "start": 352.791, "end": 368.335, "text": "Gender Index. And when I was at WHO, one of the big ones that we had was the Universal Health Coverage Index. So that's just to name a few. Now to those who who love composite indices, they will argue that they can be useful" }, { "start": 368.335, "end": 382.173, "text": "for summarizing large amounts of information, and sometimes it makes that easier. It makes that information easier to digest. Because it's compressing them instead of presenting a whole battery of separate" }, { "start": 382.173, "end": 397.63599999999997, "text": "indicators. So the argument goes, by incorporating all of these dimensions, composite indices can improve the interpretability and also make complex topics more accessible, maybe to people who aren't experts in the" }, { "start": 397.636, "end": 416.319, "text": "field. And crucially, from a more statistical point of view, it also helps us to understand. Some of the relationships between the different elements or sub indexes within the composite. Now on the hate side, the statisticians will argue with some" }, { "start": 416.319, "end": 431.46799999999996, "text": "justification that inevitably, by compressing or summarizing all of the multidimensional concepts into a single index, we lose a lot of nuance and we lose a lot of richness. And that it because of this, it may be difficult to identify the" }, { "start": 431.468, "end": 432.648, "text": "drivers of change in the composite indicator. You just see it's moving, Hmm." }, { "start": 435.92, "end": 451.368, "text": "but you don't understand why. And certainly for that reason I would always recommend publishing the the dashboard of sub indices alongside the composite so. It makes it easy for people to decompress the the composite if they want the detail." }, { "start": 453.368, "end": 465.864, "text": "Now, for this reason, when you construct A composite index, it's absolutely essential that the conceptual and methodological elements are transparent to the user. It's essential that they understand what" }, { "start": 465.864, "end": 473.408, "text": "you've done and also why you've done it so that misunderstandings. And misrepre misrepresentations can be avoided." }, { "start": 475.848, "end": 491.088, "text": "Now, one of the challenges in constructing A robust composite index is that it's not easy, and certainly it's not easy to do properly. And This is why, in a minute, Hussein will be presenting a tool that we've developed to try and help with that." }, { "start": 492.488, "end": 507.76800000000003, "text": "Now there's many instances when you compile a composite index, you will face decisions, you will face trade-offs, you will face compromises. The theory doesn't really, where theory doesn't provide a clear answer, it doesn't equip us." }, { "start": 507.848, "end": 526.264, "text": "With a clear direction, and that's why transparency is so important. Now, the reason I emphasize this is any tool like the tool we're going to see today, it can only do so much, but you still have to bring your statistical knowledge and your your your" }, { "start": 526.264, "end": 539.704, "text": "contextual knowledge to the table. Because a tool can only do so much, so you you will still always have to exercise judgment. So this tool can do a lot for you, but it can't do everything now. Nonetheless," }, { "start": 539.704, "end": 553.206, "text": "there are things that it can do now. I'm I'm just quoting from the OECD handbook now. And these are the types of steps when you're building a composite index that you should think about. So obviously you need to begin with a" }, { "start": 553.206, "end": 566.648, "text": "theoretical framework, and that framework then ensures that whatever you build is fit for purpose. So you need to be clear what you're doing, why you're doing it. And you also need to be clear about when good enough is good." }, { "start": 567.888, "end": 579.1389999999999, "text": "You then need to do an indicator selection using some defined criteria. So why are you selecting the indicators that you're selecting? And then you need to get into the messy business of cleaning up the data." }, { "start": 579.139, "end": 590.9449999999999, "text": "So imputing for missing data, all of that kind of stuff. You'll need to do some multivariate analysis to investigate the structure and suitability of the indicators, and you will also need to do a lot of" }, { "start": 590.945, "end": 604.422, "text": "standardisation and normalisation to make sure that the data are comparable. One of the most difficult choices you'll always face is around waiting and aggregation. Should you wait? If you should wait," }, { "start": 604.422, "end": 617.128, "text": "what type of weight should you use and and and at what level to aggregate? How many pillars should be in the index or how many sub sub pillars? And and how do you deal with regions? These are all really tricky questions." }, { "start": 619.448, "end": 631.036, "text": "Once you've built your index, then you need to do some robustness and sensitivity analysis to make sure that the index is stable, because there's nothing worse than producing a composite that's unstable" }, { "start": 631.036, "end": 643.968, "text": "because countries, especially if we're producing country level data, will go bananas. If they find themselves hopping around from one year to the next, so sensitivity is particularly important if you're interested in time series." }, { "start": 646.608, "end": 659.329, "text": "And then you have to do the usual stuff that you would do as a statistician. You have to do some verification against other data sources to see if what you've built actually makes sense. Now, as I said, no app can do all of this," }, { "start": 659.329, "end": 673.451, "text": "but the app that we're going to present today. I think can do some of this for you and hopefully it'll make your your life and your journey towards building your composite index a bit easier. But as I said, you cannot park your judgment at the door" }, { "start": 673.451, "end": 689.298, "text": "and just stick numbers into this this app and assume that everything will be fine. You need to exercise judgment. And bring your contextual knowledge. Now, so the purpose of today really is to introduce you to the app. We'd we'd really like to hear your" }, { "start": 689.298, "end": 701.0079999999999, "text": "feedback. As I said, it's been developed to try and help you with some of the more mechanical steps in construction of the index. So with that, I'm going to hand over to Hussein. So thank you." }, { "start": 704.528, "end": 721.2479999999999, "text": "Thank you, Steve. Like you, I have to say that good morning, good afternoon, good evening. I don't know from where you are joining, but it's a pleasure and honor to be here to thank you. I have to say first goes to Alex and also you for inviting us." }, { "start": 721.488, "end": 739.143, "text": "And for others that join us today as well. So the other thank you goes to our co-author. Two of them are from Yumtat as you mentioned Nur and also Petra and also Fernando and particularly I would like to" }, { "start": 739.143, "end": 756.368, "text": "thank you. Leila for helping up to put up this package as well. So just as as Steve mentioned, we the idea came came up here that I had challenge to make a composite index. I look at the several sources in R." }, { "start": 756.728, "end": 772.808, "text": "That there are various packaging, but each of them, as Steve mentioned, they are tackling one part. If one of them, for instance, considering how we are going to de weight or weight or scale them, the other one is." }, { "start": 773.928, "end": 789.008, "text": "Particularly focusing on PCA for example for waiting. So we could see that variety of the package there, but none of them fit the purpose as as Steve mentioned. So the idea came up to to build up a package." }, { "start": 789.728, "end": 804.3679999999999, "text": "And also looking at the various packages available to make them comprehensive as much as we can. But the other point was many of us we don't know our or we don't want to bother ourselves to use our package out so." }, { "start": 805.568, "end": 819.629, "text": "The idea was to come up with an R Irish shiny app that we can use it as an app version of that. Of course you can tailor made it for your own self because we have the package, but at the same time app version could be" }, { "start": 819.629, "end": 835.61, "text": "much easier to handle. As Steve talk about too much about why we are doing this, why it's quite important, hate and love in composite index as well. So here I'm trying to use to show the application of the package that we have provided. As Steve mentioned," }, { "start": 835.61, "end": 851.054, "text": "this is just the first attempt from our side. We are collecting feedback. To improve it, we are we are looking at the possibility to make it live so you don't need it even to run to run it on your PC," }, { "start": 851.054, "end": 866.002, "text": "but also to have a web page for that. So we are really still with other colleague. We are exploring what's the best available app. Approach that we can use as well and maybe this is opportunity that we invite" }, { "start": 866.002, "end": 884.317, "text": "for other feedback and suggestion as well. Let me share with you what we have and then through this I I would like to have a like a Q& A to hear your feedback because as Steve mentioned. Here we are to present to you this package to use it. Looking at the usage," }, { "start": 884.317, "end": 900.525, "text": "we introduced this package on the 20th of October. I could see that they are using this package more frequently than the other package that I introduced. So that's a very positive point about the usage of this package." }, { "start": 900.525, "end": 915.328, "text": "So for over three months I could see about 1000 user as well because then they are using this package which is a good sign. So looking at this, this is the look, the first look of the package, so you just need to." }, { "start": 915.608, "end": 931.408, "text": "To import your data with uploading, browsing the data, Excel or any other format as well, you can just upload your data. The first point is which was bothering me. If you have your Excel file, you want to have a header or not." }, { "start": 931.928, "end": 948.647, "text": "The first one to as a name of the Excel or not, you are able immediately to do that. So if you look at this, even we look at the very tiny point that the bother the user as well. Therefore immediately you can have a data that is here." }, { "start": 948.647, "end": 968.808, "text": "Here we use a data which is about 5. 1000 record about 194 countries that we have and you could see that the snapshot of the data as well and with some supporting information immediately you can fill out the number that you have, the variable that you have and it's fully." }, { "start": 968.888, "end": 984.608, "text": "We try to have a smart version of that and the code, not only the just to have in your our code or our package as well. So immediately the app has a scan of all your data give you the snapshot of kind" }, { "start": 984.608, "end": 1001.708, "text": "of data that you have if you have unique ID. Or you do not have unique ID, you have a report of them immediately here. This is a real data that we are using. Use this for one of the composite index which is quite complex. Having said that," }, { "start": 1001.708, "end": 1021.448, "text": "this is a time series data, but it's not. One dimension is a longitude data is a 25 years data as well. So you see that the data is quite complex with lots of indicator about the 20 less than 30 indicator here about 32 if you consider all the." }, { "start": 1021.448, "end": 1037.204, "text": "Information as well. So when you look at that immediately scanning the data is telling you based on which one you are trying to. I think that is a question. Yeah, I would say maybe just continue for now" }, { "start": 1037.204, "end": 1053.32, "text": "and when there is a good moment, maybe we can take some questions, Sure. but continue for now. Sure. So one of the things that immediately asking you, you are quite flexible to use this even it's it's giving giving you the flexibility which one you are going to" }, { "start": 1053.32, "end": 1067.456, "text": "use. Here I go for the ISO code which is a country name and then as soon as you confirm. The message will be here, so you you are OK, you you establish the first thing. You have options to use others as well," }, { "start": 1067.456, "end": 1083.225, "text": "but it's pretty much similar. So the scanning give you the which one, which indicator as they've mentioned as well. When you have a many indicator which one you are going to use, he mentioned the judgment, so you should need to judge as well." }, { "start": 1083.225, "end": 1098.992, "text": "But again we try to be comprehensive. As I mentioned, we give it chance to user and you will see later on that you have a cross correlation, you have further statistical analysis that improve your. View as well. So if you are select all of them," }, { "start": 1098.992, "end": 1116.922, "text": "therefore immediately you have this one as well. Then as Steve mentioned about the frequency of the options as well, you want to scaling one of the. Initial talk I had with Steve was that the scaling is and the direction of that" }, { "start": 1116.922, "end": 1135.463, "text": "is quite important for us as well. That was a comment that he gave me if some some of the indicator or data that you use. The maybe the higher is better, some of them maybe lower is better or some of them is mixed. Steve, do you want to comment on this direction" }, { "start": 1135.463, "end": 1149.276, "text": "as well because this is one of the practice that's is quite important. Maybe the the user they don't know that. Yeah, I think you've you've explained it well. It would be good. I mean something like unemployment. So if you take GDP as a simple comment," }, { "start": 1149.276, "end": 1163.788, "text": "let's just say GDP up is good, but then unemployment if we're using that instead of employment. So employment would be good if it's up, but if we were using unemployment. For example, unemployment going down is better." }, { "start": 1163.788, "end": 1178.4479999999999, "text": "So now we've got two indicators that are moving different directions, but they're both good and that's problematic mathematically within the index because they pull against each other. So what we have to do is normalize the data." }, { "start": 1178.648, "end": 1194.328, "text": "So we have to invert one or the other so that good means the same thing. Good in in both indicators. Either getting lower or getting higher is good. It doesn't matter which direction you do, but usually you you would pick an intuitive direction and." }, { "start": 1194.848, "end": 1211.821, "text": "So that as the index gets bigger, you intuitively understand that that's a better thing. So that's what it means. Yeah, yeah. Thank you Steve. So you could see that Steve if I if I do the text mining from Steve what's set to to to phrase here is one of them is" }, { "start": 1211.821, "end": 1229.126, "text": "normalization. So you can you can use normalize because the you you might have a different data with a different scale at the same. Time, the direction that we you see that both of them, they are here as well. Immediately when you proceed with data," }, { "start": 1229.126, "end": 1244.293, "text": "you get all the necessary information for all the indicator that you have for this case as well and also we added further information in terms of data summary. How many missing data, how much missing data that you have and" }, { "start": 1244.293, "end": 1257.5, "text": "if you are going to impute the missing data as well. Here we have a various version that you can impute the missing data as well. Having said that, when you have done this, the next is you're just pre-processing" }, { "start": 1257.5, "end": 1271.874, "text": "your data by the way. We have introduced a package that is called pre-processing. Therefore you can just before enter the data into this app, you can do pre-processing, but the package has been introduced and" }, { "start": 1271.874, "end": 1288.384, "text": "linked to that package as well, so make it much easier for you when you are running this package as well. X would be the weight. Initial weight would be we consider similar weight as well for this composite" }, { "start": 1288.384, "end": 1304.7, "text": "index as well. So you could see that everywhere is 1, but you might be interested to change the weight according to. Your wishes. Sometimes we understand that if we have a different weight might work better as well." }, { "start": 1304.7, "end": 1320.128, "text": "Of course you can do further analysis to come up with those weights, but at the same time here you have the option. If you don't do anything, you see here that. You have all the weight has been sorted out here for each of them and you could" }, { "start": 1320.128, "end": 1341.4769999999999, "text": "see that all of them with equal weight when we normalize them with goes to the index. But here you have the chance to change each weight above 1. Below one, I'm just giving you a little bit change that we can see that with a different way immediately when you do that the applying" }, { "start": 1341.477, "end": 1357.548, "text": "these changes. For instance here you could see that the process will start. You will have the other changes that it goes to the matrix and also gives you the how much positive, how much negative goes to the final" }, { "start": 1357.548, "end": 1366.248, "text": "weight. And as I mentioned here you could see that the weight now for those that they have changed. Is changing so." }, { "start": 1368.368, "end": 1386.001, "text": "These these kind of weight when you apply you go to the next one. Now the composite almost is ready to make the weight as well. So couple of countries for example here up to you top 20, you can select them," }, { "start": 1386.001, "end": 1404.432, "text": "you can play with them. We just show just here as well. You can go for the correlation analysis between those indicator that you can all play that usually we do with the statistical analysis with. Traffic has been applied here. You can make them bigger," }, { "start": 1404.432, "end": 1418.208, "text": "you can play with them, you can customize your weight. Here there are many options as you know our package and our software. They are available as well. Going to the each countries, for example here I have a country you can have a." }, { "start": 1418.288, "end": 1433.248, "text": "Different indicator or entry for the for the data as well. You give the the ranking for each year. As I mentioned here longitudinal data, you have a different weight based on what you select and here as well." }, { "start": 1433.648, "end": 1448.848, "text": "Further to that, usually we stop here, but we were thinking about it. Yes, for some case we have a time series data rather than cross-sectional data, particularly longitudinal data as well. Are we able to do that?" }, { "start": 1449.288, "end": 1469.168, "text": "So we introduce another layer to that as a. Time series analysis as well and then for different countries of course you can select this one as well, but at the same time not only the index, those variable as well. You can bring those variable just to see" }, { "start": 1469.168, "end": 1489.012, "text": "here as well if you want to do those too. If you are interested to compare them, you can select different country in order to compare them in one go with the different land service package as well. But further to that, are we able to see what's going on in few" }, { "start": 1489.012, "end": 1506.128, "text": "years time? When I'm trying to to do this, yes, you can use this as a time series analysis as well. You can have a forecast, you can have a bound, but in addition to that in terms of forecasting you are able to." }, { "start": 1506.328, "end": 1521.288, "text": "Compare your package in terms of forecasting. You might be interested. We are not implying that that's the best for forecasting, but we want it to be comprehensive that this package will or many others as well." }, { "start": 1521.728, "end": 1536.8880000000001, "text": "But at the same time, it's not only one package that you are interested. You might have like the data that we were working was very complex. So you see that is a longitudinal data over 25 years. There are many countries." }, { "start": 1537.728, "end": 1557.008, "text": "Number of indicator goes in with a different direction, with a different scaling, missing data as well. But the other point was sometimes you need to have your data with a different pillar as as we we know that it one is economy, the other one is." }, { "start": 1557.128, "end": 1576.955, "text": "Different indicator or pillar that you have. So we come up with a different addition to this package as well. We thought that can we do this in terms of say? Further analysis for that in terms of comparison of that in terms of waiting" }, { "start": 1576.955, "end": 1595.485, "text": "for the index. So it gives you the the other initiative as well in order to come up with this, then it goes to the pillar. So you might be interested. I want to have a different pillar here. I for example I introduce pillar one which indicator." }, { "start": 1595.485, "end": 1615.049, "text": "I put couple of indicator in it and then you add the pillar and then pillar two and call it B. Then couple of other indicator that you can have it as well and I can add different as I said pillar and then I just let me choose only one for instance." }, { "start": 1615.049, "end": 1631.181, "text": "So you could see that immediately when I create different pillar I have a different indicator. Iter goes with it, but we know that within each pillar we are able to change the weight as well. Currently equal weight," }, { "start": 1631.181, "end": 1646.291, "text": "but as soon as I want to have not only between pillar which is goes creating one indicator. And composited tips, but within each one I'm able to change the weight as well. You see that here if I the default is" }, { "start": 1646.291, "end": 1661.106, "text": "equal weight, but I'm able to change the weight according to what I want. Doesn't matter. Which one I select immediately when I apply the checking, the pillar will change, the effect will change and also the" }, { "start": 1661.106, "end": 1681.665, "text": "analysis as well will change as well. You are able to also do the other thing as well. So you could see that here different pillar, different weight is playing with the different composite index too. I try to to show you this. Let me just to show you further analysis" }, { "start": 1681.665, "end": 1685.528, "text": "for the pillar to make it very simple only two. Maybe." }, { "start": 1687.568, "end": 1693.448, "text": "Two pillar maybe make it easier before we opening this floor." }, { "start": 1696.128, "end": 1717.9759999999999, "text": "It's two and the C as I said we have only let's have that one to add here. But when we we do this, we create these three. They gives you the other thing as well. So you could see that in the final. I have 3 pillar with the 30% goes to final composite index." }, { "start": 1717.976, "end": 1732.633, "text": "I can see the variety of the changes, the effect and also I can see the detail analysis. If for example you see that what I choose one indicator, there is no variability, there is no index here as I'm trying to" }, { "start": 1732.633, "end": 1751.608, "text": "stop here. Because as as I mentioned, it's better we give more time to Q&A. So back to you, Alex. Thanks Hossein. Yeah for for this demonstration of the two. At the beginning your screen was a little bit stuck, so maybe some of the information has been" }, { "start": 1751.608, "end": 1752.408, "text": "Oh, I see." }, { "start": 1754.702, "end": 1772.9679999999998, "text": "lost at the beginning, but I would like to open the floor now for questions, so. Please, if you want to come in, if you want to see a particular aspect of this tool, just raise your hand and and come in." }, { "start": 1775.528, "end": 1782.608, "text": "Is anyone? Well, we have to, we have two questions online or so we Yeah. have your question Alex, which is the method we use for Yeah, yeah." }, { "start": 1784.688, "end": 1786.728, "text": "forecasting and we have a question from Khadid which was the how do we deal with Yeah, we can start with it." }, { "start": 1790.011, "end": 1808.928, "text": "data with colons? Yes. So, yeah, let's maybe start with those two questions. And I see in the meantime somebody raised his hand or their hand. So has been or Steve. So how do you deal with or how do you do the forecasting of the?" }, { "start": 1808.928, "end": 1809.3680000000002, "text": "Indicators." }, { "start": 1811.528, "end": 1813.288, "text": "What method do you use there?" }, { "start": 1815.728, "end": 1830.63, "text": "Do you want me to take it, Steve? We go for forecasting. I can, yeah, yeah, Yeah, yeah, go ahead. I can go ahead with forecasting. Here we we use automatic arimo for the forecasting, which is those that they are available in" }, { "start": 1830.63, "end": 1844.483, "text": "the R package library, but we are not bounded to only use that. We are constantly adding more and more like Walt Winter like others which is goes under the forecasting package of the auto arima as well." }, { "start": 1844.483, "end": 1862.968, "text": "But we are quite open to to add more nonparametric option particularly myself you know that my one of my. Area that I'm working is forecasting, so meaning that it's not fully aligned with what we we know, but we are adding more and more package" }, { "start": 1862.968, "end": 1877.979, "text": "to this and in future we want to have options then. For the user, like others, that you can select which forecasting method you are using. Of course Otarima use the best one, which is according to what, but we know the pros and cons of the" }, { "start": 1877.979, "end": 1891.608, "text": "Otarima as well. So that's what the first one. OK. For the second one, I leave it to you, Steve, and then I will complete. If there is any additional point to that. But the colons, I think you should answer. I don't know the answer to that one." }, { "start": 1891.608, "end": 1906.368, "text": "All right. OK. All right then. So I go for the second one as well. Well, normalization is is is a different approach as well that we know that for normalization we have to have a variety of approach that we can normalize data" }, { "start": 1906.368, "end": 1920.9019999999998, "text": "and scaling them to prepare them for the for the composite index we have used. The variety of the approach and as I said 11 what Steve mentioned, if there are negative correlation on the positive correlation," }, { "start": 1920.902, "end": 1936.917, "text": "there are two approach that we need to consider here. One is to scaling and normalize them meaning that. You we know, we know that the compositing index are very sensitive in terms of the weight of those variable that they goes in." }, { "start": 1936.917, "end": 1952.68, "text": "That's one approach which is come we can normalize them, we escape them. The other thing that Steve particularly mentioned is a direction of those as well. So there are two approach here which I you can see from your question." }, { "start": 1952.68, "end": 1971.078, "text": "One is a scale and the magnitude if you say per statistics for the variable that we have for for instance the GDP and you look at the number that is the ranges from 1 to 10 and billion to. Million to billion. So that's tool that we can normalize the" }, { "start": 1971.078, "end": 1989.814, "text": "scale. The other approach that SD was commenting as that is the the approach of the direction of those, for example if they are. You have 10 variable, five of them are negative correlated, five of them are positive correlated as well. How you are dealing with that?" }, { "start": 1989.814, "end": 2006.507, "text": "Because if you make an index some of them, so we have an approach in the in the tools that you can choose which one. You need to select or deselect or mix of them. You could see that when I was playing with that we have above down or mix of" }, { "start": 2006.507, "end": 2021.622, "text": "them that you want to to mention again your example listed maybe here just to make it clear. Well, I was gonna, I see Petra's online. So Petra, maybe do you have an example from the inclusive growth index?" }, { "start": 2021.622, "end": 2027.568, "text": "And I'm putting you on the spot, but you've had to do this a lot with the IGI." }, { "start": 2033.808, "end": 2042.4479999999999, "text": "OK, she's not playing, so it it looks like she's online. So Petra was one of the the constructors of this, but she she also has a lot of expertise in building composites. Yeah." }, { "start": 2044.961, "end": 2046.208, "text": "So it's a pity I thought she she might have had a. Yes." }, { "start": 2048.488, "end": 2066.76, "text": "A better example than I had. So she could come in because I switched on the cameras and the microphone. But yeah, maybe there's an issue. Yeah, so, Uh, OK, OK. but you basically already kind of answered then now the question of Louis" }, { "start": 2066.76, "end": 2077.688, "text": "Arturo, right? Because he was asking about the normalisation and the negatively correlated variables. But maybe in the tool Jose, maybe you can show it once more where where to find that because the screen was" }, { "start": 2077.688, "end": 2079.808, "text": "Yeah, yeah." }, { "start": 2082.044, "end": 2089.248, "text": "a little stuck at the beginning and. Oh, I see. I see. I see. It wasn't really a we weren't really able to follow that. So maybe you can show that once more and then we can move to other questions. Yeah." }, { "start": 2091.208, "end": 2106.608, "text": "Yeah, yes, here it is one of those. You know the the question that that mentioned you can look at the correlation, you can you can look at the further analysis here as well in terms of to find out which one are more." }, { "start": 2107.048, "end": 2124.181, "text": "Say correlated or they are not correlated as well, but at the same time when when we look at look at the tools, you can import and then do further analysis as well on the correlation side too. So that's basically give you the the idea" }, { "start": 2124.181, "end": 2139.246, "text": "how you should play with that. But if I then the general approach you can do. But here we use the other alternative approach like PCA that is embedded. You can use it because therefore you don't have those issue on the larger" }, { "start": 2139.246, "end": 2140.768, "text": "scale as well. But these tools that we just wanted to OK." }, { "start": 2142.735, "end": 2143.9280000000003, "text": "show here as well in addition to PCA that. Yeah." }, { "start": 2146.048, "end": 2155.888, "text": "Inherently consider those correlation as well. You can have a visualization to as well to take it here in addition to other one as well. Alex, there was a question as well for the one Yep." }, { "start": 2159.245, "end": 2165.168, "text": "of Russia, I think. Yes. Can we go to Russia now? You have your." }, { "start": 2167.928, "end": 2185.6879999999996, "text": "Get there and. Yeah, thank you. Thank you for this opportunity and for the food for presentation. I just like have a few questions. Maybe I was not like able to catch up for like I not just. I just want to make sure for some things it as far as I understood this is an R" }, { "start": 2185.688, "end": 2203.888, "text": "package. So if I if I am able to download this R package, I will just like able to use the the composite index tool, right? This is for question one I want to ask regarding the data. So this tool basically applied to any time series data that I have." }, { "start": 2203.888, "end": 2203.928, "text": "Oh." }, { "start": 2207.248, "end": 2225.568, "text": "Yeah, boys. For anytime series data and my question #2 was regarding the weights. So I would I only I would only I have to upload the data and the system and this automatically compute for me the weights. I only like just need some a bit of." }, { "start": 2225.568, "end": 2242.2079999999996, "text": "Mm. The clarification regarding the PCA based weight and yeah, finally I just have, I had the question regarding the forecasting, but thank you, I got the answer from Hussem. So that's it. Thank you." }, { "start": 2245.528, "end": 2264.728, "text": "Yes. So so thank thank you for the question. Yes, I just post here in the chat the link to the R package. We just need to have a line few, I mean one line you just you write it and the package is already available in." }, { "start": 2265.168, "end": 2284.627, "text": "So you just need to to add it to your code immediately. When you run it, you will have the option that the same app will come up and therefore you just add your Excel file as well by the way. By the way, we try to to make the user user friendly" }, { "start": 2284.627, "end": 2300.0480000000002, "text": "app, meaning that as soon as you upload the data is considered all the issue that you might have. But I'm sure everyone has a different rules like one has a comma, the other one doesn't have a comma, you know one." }, { "start": 2300.648, "end": 2314.937, "text": "One has a title, what the other one has a doesn't have a title on in the Excel file or the file that they have. We look at those issues immediately. You can use it, you can upload the data. The second question to ask that's for the app." }, { "start": 2314.937, "end": 2332.14, "text": "The second one question was the weight. As as I mentioned on the indicator, immediately the the first thing that the app it does, it goes a smartly scan your data, number of variable, number of things, everything that is automatically" }, { "start": 2332.14, "end": 2349.288, "text": "indicated in the app. Then the predefined is equal weight, but as I mentioned and I play with that, you can change the weight yourself as well, particularly after the further analysis that you have, therefore you are able to to change the weight." }, { "start": 2349.288, "end": 2370.136, "text": "The way that you would like to have. I knew that for many of us for many user using the weight that we believe is better weight and to make it maximum, say variability or minimum in terms of. Arrow would be optimal for us as well. So you are able to change the weight" }, { "start": 2370.136, "end": 2384.66, "text": "yourself as well backing to the other issue like the PCA or any alternative method that we can use it for the weight. Yes, that's alternative approach that can be. Used to. I think these three question I could grab," }, { "start": 2384.66, "end": 2402.316, "text": "but if I missed anything, just write it in the chat. I come back to that and thank you for that. Thanks, Jose. There's also a question from Oscar Martinez. Oscar, you want to come in live? Please then unmute yourself and maybe put" }, { "start": 2402.316, "end": 2407.048, "text": "the camera on if you can. Otherwise I can read it out. Oscar." }, { "start": 2414.048, "end": 2416.0480000000002, "text": "OK, but I can read it out. Oh. Right. Sorry, no, just Jeffrey that I is. I think that the question is Yeah." }, { "start": 2420.897, "end": 2435.0480000000002, "text": "straightforward is how to if the tool have some features to make you easier to decide which indicator to select from another if it's highly correlated. Pictures up." }, { "start": 2441.048, "end": 2451.888, "text": "Yeah, yeah. Product and gross national income, which are highly correlated if the tool can help you decide which one to use." }, { "start": 2454.088, "end": 2468.454, "text": "You go ahead and save and then I complete. Yeah, I was gonna say, I mean this is an excellent question and and this is one of the challenges of composites is is redundancy. There's nothing I don't. I think I'm correct in saying there's" }, { "start": 2468.454, "end": 2482.047, "text": "nothing explicitly in the tool that actually helps you. So you you can look at correlation, but. This is what I was saying. You still need to bring your own judgment. You'll still have to decide for yourself which is the best indicator." }, { "start": 2482.047, "end": 2495.119, "text": "And that's not going to be a a technical decision. That's going to be really context specific. Uh, what do you want to do? Why are you constructing the index? And it may come down to other issues, like even though the two are highly" }, { "start": 2495.119, "end": 2506.768, "text": "correlated, maybe one just you have to impute less, so you might decide that the data quality is better and one or it has better country coverage. So there's a range of issues that will help you to make the decision." }, { "start": 2507.208, "end": 2510.928, "text": "Um, but I I think that's really where human judgment comes in." }, { "start": 2512.968, "end": 2530.048, "text": "Yeah, thank. Thank you. Sebas. Perfectly you answered the question. I would like to add 2 points here. So as you saw that we we added a couple of feature and based on the comment that I received, it's encouraged me to add more feature in terms of showing that the correlation." }, { "start": 2530.568, "end": 2545.83, "text": "A positive, negative or how contribute to those as as well. But at the same time you saw that we added other feature as well. So when you you have the chance on the option that you can downplay or amplify those indicator that you would like to" }, { "start": 2545.83, "end": 2563.528, "text": "play with that when it goes to pillar for example this is a complete. Example that maybe those are yes correlated, but they go under different pillar as well. You might see that this is not good disadvantage for the composite index," }, { "start": 2563.528, "end": 2579.215, "text": "but at the same time you. Need them for your pillar or sub indicator that is goes to composite index as well. So while you can keep them to make the composite for one of those pillar that later on goes to final composite index," }, { "start": 2579.215, "end": 2595.128, "text": "but at the same time you might play to bring down the weight when you make it. The final composite index as well. If I may have the chance to answer one of this question as well, which they asked, did you have you played with the real" }, { "start": 2595.128, "end": 2611.649, "text": "data or verified them? Yes, that's the thing that immediately we do that when we making composite index, we did it for the other data as well. Which also did this package has been featured in UNC as you saw Alex as well." }, { "start": 2611.649, "end": 2629.866, "text": "But at the same time we try to use a complex data set as well and that's as at the beginning of this call my appreciation went to. To our colleagues from that Petra and new that help us to use their data which is a" }, { "start": 2629.866, "end": 2646.853, "text": "quite complex. So I I recall that the Steve said if your index or what we are presenting works for though that data definitely can work for other simpler data as well and I can confirm. And that the data now I just show is works for dead data." }, { "start": 2646.853, "end": 2651.608, "text": "I use dead data as well 2024 data set as well, which is quite available as well." }, { "start": 2654.968, "end": 2678.962, "text": "Great, Jose. So this answers also the question by Amro in the chat. There's another question by Michael James. Jacob Jacobs, you want to come in? Otherwise I can also ask it myself. All right. Uh. OK, so he asks, what options have you already considered" }, { "start": 2678.962, "end": 2699.957, "text": "for making the app available as a website? Are there any privacy issues to be aware of? I can answer this as well. Well, the ultimate aim was to have these tools available on one of the web page of the one of the international organization or" }, { "start": 2699.957, "end": 2716.273, "text": "some other data representation that we can use. As I mentioned, we are looking at the. The best option because it's not only to have the current version, but we keep updating it. So what's the best option is to have it." }, { "start": 2716.273, "end": 2732.5429999999997, "text": "But still we can't. There is no privacy issue. There is open source. That's why we give the code to everyone that they can use it. No privacy issue. And is is available and the intention was to make it up to make it even much easier" }, { "start": 2732.543, "end": 2748.168, "text": "for the user to use it rather than running the our code. But that's the next step that we are looking at currently, which organization or platform to use to the wide range of applications. For these tools as well." }, { "start": 2749.128, "end": 2749.808, "text": "But we are quite open. We are quite open to what suggestion. Yep." }, { "start": 2753.928, "end": 2766.328, "text": "Maybe the question was also referring to the data which is being uploaded. Let's say if the data that you upload has some privacy issues, No. if then if you put it on a website, if it is kind of protected." }, { "start": 2766.328, "end": 2778.288, "text": "Yeah, yeah. OK, there is there are two questions also by Marcelo Lafleur. Marcelo, you wanna come in?" }, { "start": 2783.688, "end": 2802.897, "text": "I think Petra also answered the the question in the chat. Yeah, Marcelo, if you cannot, I can also read it out. So Marcelo has two questions. So he says thank you for this. It is much needed and much appreciated. I may have missed it," }, { "start": 2802.897, "end": 2823.248, "text": "but does the app provide some analytics for example in the? PCA, does it give you the fit statistics? That was one question by him. And the other question was, is it possible to have different weighting methodologies for different levels of aggregation?" }, { "start": 2823.608, "end": 2840.672, "text": "For example, using PCA for the sub indices and equal weights for the pillars. Thank you. So I will go for the waiting. That's a cover basically boost. Yes, as I mentioned, one of the beauty of this tools that we" }, { "start": 2840.672, "end": 2855.727, "text": "provided is you can have a different weight for the different pillar as well when you created as well. You could see that when I was running I give. What two? I created three pillar, one of them I changed the weight," }, { "start": 2855.727, "end": 2870.721, "text": "the other I fixed so I didn't change as well. So you can play with that and how the the weighting based on the which criteria goes. We know that we have a different. Approach provided. We can use a variation variance," }, { "start": 2870.721, "end": 2886.8880000000004, "text": "we can use a PCA, we can use others as well. The good news as is that we are adding more statistical measure goes for the weighting as well. We are expanding as as as Steve mentioned, that's the first step we look at what's." }, { "start": 2886.928, "end": 2902.9280000000003, "text": "We have available in the packaging either R or Python or anything else and what's the demand is called for this composite index and we thought that this could be the first step as Steve mentioned in the in his remark." }, { "start": 2903.288, "end": 2918.928, "text": "It's not comprehensive as everyone wishes to have it, but I think we are moving toward making it better and better and such a this opportunity we take it to improve for the better usage of this apps for all." }, { "start": 2920.488, "end": 2924.2079999999996, "text": "Yeah, that's really great. Um, any other questions?" }, { "start": 2927.688, "end": 2931.5679999999998, "text": "There was one question by. Are you going to share the R code as well? But." }, { "start": 2934.621, "end": 2949.1279999999997, "text": "But I guess it's yeah, it's online, it's open source, right? So that's here. Um, Marcelo has another question. If you wanna come in now, it's OK if not, uh." }, { "start": 2951.328, "end": 2970.319, "text": "Yeah, so he's basically asking where can we contribute to the project or provide feedback? He says he's on the train. He cannot come in. Anywhere you want, Marcelo, e-mail us. Probably the easiest you can get." }, { "start": 2970.319, "end": 2978.528, "text": "We can get. You can get our contact details, but yeah, we're happy to receive any feedback. Yep." }, { "start": 2980.728, "end": 2995.888, "text": "That is great. Amber was mentioning, yeah, we can discuss maybe to put it on one of our websites or service here at UNSD in the data innovation and capacity branch. So yeah, let's have a discussion on that." }, { "start": 2996.008, "end": 2997.368, "text": "I have I have maybe one question also therefore the composite indicator as Yeah, that would be good." }, { "start": 3002.131, "end": 3022.781, "text": "Steve mentioned it at the beginning somewhat that when you construct composited indicators you also need or should do a sensitivity analysis of the composite indicator. Does the tool already support this or is it planned for the future?" }, { "start": 3022.781, "end": 3026.688, "text": "Maybe you can say something about this?" }, { "start": 3031.208, "end": 3045.3349999999996, "text": "Yeah, I mean, I guess this is where, I mean this is something we could look at, but to do that you'd have to import in the other data. Well, it depends what type of sensitivity you're doing if you're comparing it with" }, { "start": 3045.335, "end": 3057.368, "text": "other data. You'd have to import that data, but for the normal sensitivity analysis, it's probably something we haven't done enough of yet, but that for sure that's one of the areas I think we need to strengthen but." }, { "start": 3058.328, "end": 3068.2079999999996, "text": "That would be the idea that you can, especially for time series, um, that we need to do some more kind of stability testing, but there are basic stability tests there already." }, { "start": 3070.848, "end": 3081.968, "text": "OK. Yeah. Then I saw a question by Danielle. Danielle, you want to come in, Danielle?" }, { "start": 3084.528, "end": 3109.145, "text": "Hello, yes, thank you and Congrats for the tool. Just a very simple question. One of the typical cases of composite indices being sidetracked is a few outliers driving correlation to strongly. So it's a a standard practice to cut some of the outliers or or or apply some kind" }, { "start": 3109.145, "end": 3121.408, "text": "of a treatment of variables that are very skewed. Think of log transformation or so. I was wondering whether this is part of the. The tool or is it foreseen? Thank you." }, { "start": 3125.928, "end": 3147.299, "text": "I can answer this question and also the another one. Thank you. Thank you for the question Daniel. This is as as you mentioned is quite important aspect as well and therefore as as. Mentioned during my presentation, we have a different package we introduce" }, { "start": 3147.299, "end": 3161.528, "text": "as a pre-processing. So in the pre-processing you know that all of this it goes under the pre-processing of the any data that we have and in that particular package we include all of this pre-processing and." }, { "start": 3161.968, "end": 3175.533, "text": "Preprocessing package is one of those package that we introduce. It's not the part of this package, it's a different package that is freely available in R and we have lots of approach that dealing with that." }, { "start": 3175.533, "end": 3189.5809999999997, "text": "In fact we use preprocessing with record linkage and you know that. We have a big issue when we link to database pre-processing is one of those outlier log transformation and many others as well. So that package." }, { "start": 3189.581, "end": 3208.161, "text": "But having said that, when you are talking about this, we might bring some of those elements here as one of the top. That may be to have a different preprocessing as well. Here we have, but maybe we can highlight it as well. So which method may be user data level" }, { "start": 3208.161, "end": 3222.965, "text": "rather than to having automated approach for preprocessing as well. And thank you for that suggestion. We will make as a new tab for to highlight as well and thank you for that. And I I would like to use this" }, { "start": 3222.965, "end": 3240.038, "text": "opportunity now Daniel asked for this what we are currently doing this. Composite index to bring another element where that whenever possible to link it to GIS data as well. So in the recent application that we had" }, { "start": 3240.038, "end": 3256.901, "text": "the final output composite index, we may link it to the GIS as a map. And you could see that we we mapped the index as well with the. With the help of our colleague as well. So we are trying to include include now" }, { "start": 3256.901, "end": 3270.866, "text": "as a GIS to this package. That's a new addition that we are going to add to this package as well. There are other question. I'll wait for you Alex to read them. Yeah, there's one more question by Christoph." }, { "start": 3270.866, "end": 3289.162, "text": "Christoph, you want to come in? If you can. Hello, good evening and thank you for this presentation. Just I was wondering when we are experimenting with some different weights, whether there would be a possibility to" }, { "start": 3289.162, "end": 3304.984, "text": "visualize the different indexes you you had in mind and have some sort of history and compare. The current version with a new one with different weights. If you have thought about some sort of, you know," }, { "start": 3304.984, "end": 3326.775, "text": "keeping the history of what you were doing to to better adjust your your favorite index somehow. And after that maybe I will have a suggestion for hosting this this app on posit or R studio directly on a studio that is posit now hosting a web page" }, { "start": 3326.775, "end": 3341.777, "text": "where you can you can really have. Updated version anytime you want on this tool, but that that cost a little bit of money. Of course it's not free. Yeah, yeah, Yeah, I know, I know, I know. but but still it's a very convenient because you can update anytime you you" }, { "start": 3341.777, "end": 3345.248, "text": "want to push that. Yeah. But anyway, thank you. Yeah. Thank you for your presentation. It's very interesting. Yeah." }, { "start": 3349.048, "end": 3368.15, "text": "Yeah, thank, thank. May I, Steve to answer this and that you can. Well, no, thank. Thank you very much for this. In fact, it's a very interesting suggestion when I was reading your, I mean text. In terms of question, I really enjoyed it because that gave me" }, { "start": 3368.15, "end": 3384.48, "text": "another layer. You know why? Because while we have the weight, you are able to change the weight or you are able to look at the pillar and but it doesn't give you can you you currently. You can do it by sub pillar, but if you use all the pillar you don't" }, { "start": 3384.48, "end": 3399.4480000000003, "text": "have a history that you want to look at the final composite index. So this is very good and that I will make a note to make another one for that another tab that we can compare different composite index what we." }, { "start": 3399.528, "end": 3415.314, "text": "Variation if you change the weight and then the other thing is that we might be able to add a little bit more simulation or bootstrapping onto that as well to see that how it works and how give us the flexibility and I I really welcome your" }, { "start": 3415.314, "end": 3429.0080000000003, "text": "suggestion Albert for that and give us another addition. Thank you for that, Kirsten. And also if I look at the the cost, I already look at it as well. Depends on the number of users, of course. Thank you for that." }, { "start": 3430.088, "end": 3440.208, "text": "Yeah, thanks. So we still have one or two minutes. So one more question maybe we could take if there is any." }, { "start": 3442.808, "end": 3459.9680000000003, "text": "I don't see a hand. I don't see any other question in the chat. I hope I didn't miss any. Otherwise you can still reach out to Steve and Jose afterwards and and yeah, ask them your questions and so on." }, { "start": 3462.008, "end": 3479.0879999999997, "text": "So if there's no further question, then I would like to ask everyone. Maybe you can switch on your camera for a few seconds, maybe also your microphone. Now everybody can unmute and give our speakers a big round of applause." }, { "start": 3479.808, "end": 3496.1259999999997, "text": "OK. Thank you. Thank you. Thank you. Thanks, Hussein. Thanks, Steve. Thank you. Thank you. Thank you. Thank you. Thanks everyone and thanks for your comments and suggestions. Very much appreciated. Yeah, I would like to thank the first to our" }, { "start": 3496.126, "end": 3498.0080000000003, "text": "host, Alex and you for inviting us and also to Thank you." }, { "start": 3499.926, "end": 3518.22, "text": "those participated and also to our coauthor that are some petrals here, some of them like Layla. And Fernando, I know they are not here, but contributed to this task as well. As we mentioned, we take all the suggestion and that you" }, { "start": 3518.22, "end": 3537.648, "text": "have and we try to improve the tools as well. Yeah, thanks everyone for joining. In the chat I have put two links, so one to our LinkedIn group. That's basically the global network of data officers and statisticians has moved to LinkedIn." }, { "start": 3538.008, "end": 3554.888, "text": "Last year and also a link to subscribe to our newsletter. So please follow both links and sign up and join. And thanks again to Josein and Steve for the great presentation. And yes, Josein, you still want to say something? Yeah, please." }, { "start": 3554.888, "end": 3571.158, "text": "Well, sorry, yes, sorry. Before we go, you know and me and Steve, we are heavily involved in the Journal of Official Statistics. So I would like to use take this opportunity to invite those that they were all of you, you're working on the official statistic." }, { "start": 3571.158, "end": 3587.545, "text": "This is a venue for all of you to publish your high. Quality paper. Here we are to serve you. Me and Steve, we try to improve the journal as much as we can and we need your support to submit the high quality paper and hopefully we can take it to the next tier as well." }, { "start": 3587.545, "end": 3610.128, "text": "And thank you for that as well. Yes, thank you Jose for mentioning this as well. So please be involved in the journal and thanks everyone for joining. Have a great rest of your day. And yeah, please subscribe to our newsletter and go to our LinkedIn group. Thanks everyone." }, { "start": 3610.808, "end": 3614.768, "text": "Bye, bye. Thank you. Goodbye. Thank you. Thanks, colleagues. Bye." } ] }