 Hello and welcome to today's Gateway to Global Aging Data Advanced Workshop. This is a joint webinar of the UK Data Service at the University of Essex in the United Kingdom and the Center for Economic and Social Research at the University of Southern California in the United States of America. My name is Berate Lichtfahrt and I will demonstrate how you locate and access data on aging via the UK Data Service in general and the data of the English Longitudinal Study of Aging ELSA in particular. I will then hand over to Dristin Phillips, Project Manager for the Gateway to Global Aging Data, who works for the Center for Economic and Social Research. In his talk, Dristin will briefly introduce the Gateway to Global Aging Data and give a detailed example of cross-country analysis using the Harmonized ELSA and the Harmonized Health and Retirement Study data. Okay, let's start with how one goes about locating and accessing data on aging, slash ELSA data via the UK Data Service. The roadmap for my talk will be as follows. First I will give a brief overview of the UK Data Service. I will then talk about how to find access and explore data and finally mention user support options and resources available. What is the UK Data Service? The UK Data Service is a comprehensive resource funded by the ESRC. It is a single point of access to a wide range of social science data, as well as the data. We also provide support, training and guidance. Please have a look at our recorded webinars at the URL provided on this slide. That is our website. Now who is it for? It is for academic researchers and students. It is for government analysts, for charities and foundations, business consultants, independent research centers and think tanks. Our data sources are official agencies, that is mainly central government like the ONS. The data come from international statistical time series, but also from individual academics who hold an ESRC research grant, for example. The data come from market research agencies, public records, historical sources and we also have access to international data via links with other data archives worldwide. The types of data collections we hold are survey micro data, aggregate statistics, census data and qualitative and mixed methods data. The kinds of data we hold are large scale government funded UK surveys, longitudinal surveys following individuals over time, international macro and micro data, census data, business data and qualitative and mixed methods data. Today I'm focusing on longitudinal data and out of this list of key longitudinal data, for example, I'm focusing on the English Longitudinal Study of Aging or in short ELSA. The English Longitudinal Study of Aging is a longitudinal survey of aging and quality of life among older people that explores the dynamic relationships between health and functioning, social networks and participation and economic position as people plan for, move into and progress beyond retirement. We hold so far seven waves from 1998 to 2015 and as we speak wave eight is underway. It started in May 2016 and will be conducted until June 2017. The study number in our data catalog is, stands for study number 50-50. So now why is this so interesting? Well, one in three people in England are now over 50, which means it's really important to understand what life is like for England's aging population. And ELSA helps the government plan for health care services and pension systems to best meet the needs of this growing population. The main objectives of ELSA are to construct waves of accessible and well documented panel data, to provide this data in a convenient and timely fashion to the scientific and policy research community, to describe health trajectories, disability and healthy life expectancy in a representative sample of the English population aged 50 and over, to examine the relationship between economic position and health, to investigate the determinants of economic position in all the age, to describe the timing of retirement and post-retirement labor market activity, and finally to understand the relationships between social support, household structure and the transfer of assets. The English Longitudinal Study of Aging collects data from a representative sample and that's about 10,000, so it's more or less, it varies and people have to be, response have to be added to the sample on particular waves, but roughly about 10,000 of the population aged 50 plus in England on a range of indicators such as health, economic circumstances, well-being and social participation. One of the key findings of the most recent wave was, for example, that more people aged over the state pension age are now working than ever before, over a third of 60 to 69 year olds were either employed or self-employed in the last month. Now that can be a good or a bad thing or both. It can be a bad thing because it might mean that receiving a pension is no longer sufficient, is no longer a sufficient income. However, it could be also a good thing as it means older people are fitter and are able to work longer. In any case, there might be a good aspect to that which is a second finding and that is that the majority of people over the age of 50 report hardly ever or never experiencing feelings of loneliness. In wave three, people will also ask thinking back over your life with its wide variety of enjoyable as well as difficult experiences. Please write about three aspects of your life that have been especially important to you and how they affected you. That is part of the life history interview in wave three and that comes as a particular text file with your data. Now how would you go about finding Elsa data in our catalogue? You have three options basically. One is to use the key data search. Here you would click on get data, then on the left-hand side select key data and you find a similar structure shown you in the previous slide and you would click on longitudinal studies and then scroll down the screen and you would find the Elsa data. Another possible option is to use the data catalogue search and you would then select the type and specify data collections and then also type the title and you would find it and the third option is to search by theme. Again you would select get data and then on the left-hand side data by theme and you can see we have a couple of themes and one is aging. You would click on it and again there would be a table with relevant data sources, one of which would be the Elsa data. How would you access Elsa data? It is web access to Elsa data and here's a URL so you would just need to click on that and follow some instructions I'll give you in a minute. Just a word before that. It requires agreement to special conditions online. It's not a problem. I will also explain what the special condition is later on and you can click on it online and it will not delay your download process. The data are supplied in SPSS, data, SAS and TAP and RTF for the life history essays. So this is how you download the data and the documentation it all comes as a zip bundle. In the data catalog you find the study 50-50 on that particular URL I have given it here and then circled in red is what you need to click on its download order and then you will be asked to register if you haven't registered with us before. You will need to provide a couple of details about yourself, your institution and the project you intend to use the data for. And then in the process of that you will also ask to accept the special conditions. You click on that and then basically it shows you that little box summarizing the special conditions and it's mainly to do with confidentiality. So you have to agree not to link or attempt to link the Elsa with zero data to the health survey for England data because they are a sample actually taken from the HSE. Not to agree not to use the wave zero data in any way to identify participants from Elsa or HSE and finally to agree not to use nor attempt to use Elsa data to identify specific geography from which the study sample was selected nor to claim to have done so. So you accept that and then you choose which format you would like to download your data in SPSS data or else. If you cannot remember that you just go online and every step is outlined under get data and how to access data and step by step you can follow what to do in order to access data. Now exploring Elsa data online we have a Nesta browsing and analysis system available and it allows users to search for locate, browse and analyze and download a wide variety of statistical data within a web browser however the Elsa data are too big and also you have to accept special conditions but there is a teaching dataset available and you can actually browse it so unless you want to do cross tabulations and regressions you don't even need to register for that. You can just go and browse the data and see frequencies but as I said if you want to do cross tabulation or regression then you will need to register again. That's online, it's straightforward and not problematic. So that is a screenshot from Nesta and here I have used the Elsa teaching dataset to just give a very quick example what for example you could look at. The question here was on balance I look back on my life with a sense of happiness and what did people answer to that? Do they look back with a sense of happiness or not? And that's actually quite a good result about 95% of the Elsa respondents on balance look back on their lives with a sense of happiness. You can then also do graphs and all sorts of things and download this. And finally I would like to say something about our support and resources available online. We provide video tutorials and webinars also this webinar will be available afterwards. We provide case studies so you can get inspiration of how other researchers have used the data. We have guides, we have themes as I showed and also teaching data and other resources on teaching and we have a help desk which answers individual user questions. So this is where you would find the webinars and then use and events. You would find the case studies under use data and data in use and here you could also search for which case studies do we hold based on Elsa data. And teaching resources you find under use data, left hand side teaching with data and then whatever you need to choose from the drop down list. FAQs is quite a useful source of help otherwise please get in touch. These are our contact details and we are also on Twitter, Facebook and YouTube. Now if you have any questions, please type those into the box on the right hand side of the screen and I will answer your questions after Dresden has finished his presentation or depending on the time via email following today's webinar. Now I'm handing over to Dresden for his talk on the Gateway to Global Aging Data and on aspects of cross-country analysis on aging using the Harmonized Elsa and the Harmonized Health and Retirement Study data. Okay. All right thank you, so I think we'll just transfer over to my screen. All right thank you. So one of the files that you will download, one of the data sets that you'll download from the UK data archive as part of the ELSA data is the Harmonized Elsa and this is a data set which is produced at my center here at the University of Southern California. So just briefly, the Harmonized Elsa was created to provide ELSA variables which are harmonized to be comparable to other international health and retirement surveys. There are health and retirement surveys which are designed to be comparable in how they ask their questions and who they ask their questions to all around the world in every single continent and there's more than 30 countries that have a kind of similar study. So the Harmonized Elsa currently incorporates the first six waves of ELSA. So starting in 2002, all the way through their wave six which is in 2012, we will be incorporating wave seven probably over the next couple months as that data was just released. The data set is structured in fat format. So in the original ELSA data that you would download, you get kind of one data set for every wave but the Harmonized Elsa is just one data set. So every individual is one record and then the different wave reports are listed out and we do that using a variable name and convention. Variables are defined as similarly as possible to the RAND HRS definition of a variable. And if you're not familiar, the RAND HRS, which I'll be using in the example today, is a kind of a harmonized, cleaned version of the health and retirement survey, which is conducted here in the United States by the University of Michigan. And the RAND HRS is this version, this very cleaned, user friendly version of the HRS data that, you know, primarily something like 90% of the research that is done using the HRS usually starts are also involved using this RAND HRS data. And there's a couple hallmarks of the RAND HRS data which we incorporate into all of our harmonized data sets. So one is that variable names use a really simple naming convention. So a variable like R1 work is whether the respondent is currently working in wave one. So R for respondent, one for wave one of ELSA. So this would be the 2002 wave, the baseline wave of ELSA. And whether the respondent is currently working. And variable names also indicate how similar to the RAND HRS version of the variable is. So for example, a variable in the harmonized ELSA called R1LBRFE is the labor force scale in ELSA. And labor force and how labor force questions are asked are really specific to countries. So the scale that's used in ELSA is quite different in the HRS. In that case, we don't give it exactly the same variable name. But we give it this ELSA specific variable name by using the underscore E at the end of the variable name. Which is to say that we still include that information because of course it's very important to have labor force status for ELSA. But it's not going to be exactly comparable to the HRS data. We also include spouse versions of most variables. For example, a variable called S2 work is whether the respondent spouse is currently working in wave two. These are really helpful if you're interested in looking at family effects. It saves our researcher a lot of time of not kind of doing spouse matching to look at things. Which are quite similar to the NFI for instance. Households where both people are working. One person is working or no people are currently working. And the harmonized ELSA as well as all harmonized data sets are accompanied by a code book. Which includes a lot of documentation for exactly how these data sets were created, how each variable was derived, what are some differences that might have happened between different waves of ELSA, for instance, that could have changed the variable. And how is this variable perhaps different from the variable in the RAND-HRS which we kind of use as our basis for comparison. So we're going to do just a really simple cross-country analysis using the harmonized ELSA data and the RAND-HRS data. And this is just going to be a very quick way to you to see some of the advantages of using harmonized data, especially when comparing between different countries. So we're going to ask this question, are there gender differences in cognition? And are these differences consistent across all age groups? And then do we see the same pattern in England as we do in the US? And this is coming off some work that was done on the Chinese Health and Retirement Survey which is called Charles. So here are the steps we're going to do first. First, we want to identify relevant variables in the harmonized ELSA and the RAND-HRS. We're going to create a pooled or sometimes called a stacked data set with variables and observations from both harmonized data sets. We're going to have a few prepared variables, we'll apply weights, and we'll analyze cognition across genders and ages for each country. And I'm going to be conducting this webinar and you'll see in STATA today. I've also provided you the program that we'll use as a handout. You can see in the handouts, there's a handout called advancedELSAwebinar.doc. This is actually a STATA program. If you just remove the last C there on the file extension, you can follow along with me if you would like. But I'll also show everything on screen so you don't have to. Let me show you our website. So we've built a website. This is the Gateway to Global Aging Data. It's available at g2aging.org. And it's an overview of all the studies which are conducted around the world. So you can see here many of the studies and the countries they're conducted in. And I'll briefly mention that we have this tab here, a download tab, which also has links to all the survey data and how to obtain the harmonized data. So, for instance, for ELSA, we direct you to the UK Data Service. For the HRS or the RAND-HRS, we're going to direct you to the University of Michigan, which is where you would sign a very similar user agreement to get access to the RAND-HRS data and the original HRS data to download this. These studies are all part of a network. And part of being part of that network means that you release all your survey data available for free for researchers around the world. So all of these data are free and available to anyone doing research. So first, let's talk about cognition. So cognition is kind of a multi-dimensional, many parts of cognition. So we're going to be looking at a verbal memory part of cognition. And this is a test that was conducted both in ELSA and in the HRS. And it's very comparable. And it's a word recall test. And there's two parts to it. So the respondent is given a set of ten words. This is an example from ELSA. Not everyone gets the same word list. But this is a ten-word list. And the interviewer says to the respondent, I'm going to say ten words for you or a computer is going to read ten words to you. And then I'm going to ask you to recall them back to me. So the computer or the interviewer says the words are hotel, river, tree, skin, gold, market, paper, child, king, and book. I'm going to give you a certain amount of time. Can you read these back to me? So that's the first part of the test. The second part of the test is that a later time in the survey, the interviewer is going to say, remember those ten words that I gave you. How many of those can you recall back to me now? So we develop a score, just a word recall score. And it's possible that someone gets all 20 words right. This is very hard. I would certainly not be able to do this. This is part of a long test. So the top score would be 20, meaning you remembered all ten words, both right after they were said to you and later in the survey. And the lowest score would be zero, meaning you couldn't recall any of the words immediately or lower in the survey. And this is a common kind of verbal memory cognition test, which is used in a lot of these surveys, because you really do see changes in different age groups in the ability to remember these words, both immediately and delayed. So we want to identify relevant variables. First, we'll talk about the harmonized ELSA. So for this analysis, we're going to not be using necessarily the longitudinal aspects, because it's just a quick analysis. But we're going to be kind of doing a cross-sectional analysis for just 2010, which is ELSA way of five. So we're going to need a measure of cognition. We're going to need something to identify gender, because we're going to look at if there's a gender difference. We need to identify age. So we need an age variable. And then we also want to weight the survey. Most of these surveys are conducted with complex survey design. And they sometimes include over-samples of smaller but important demographics. So it's really important to use analysis weights. So if you wanted to find these variables for yourself, I'll show you two ways to find them. And the first is at our website, the Gateway to Global Aging Data. So we built kind of a survey concordance that allows you to find variables between different surveys. So for instance, we know we're looking in the harmonized ELSA. And at this point, we can also say we're using the harmonized ELSA and the RenHRS. We can select a year or sets of years, but we're looking at 2010 in both surveys. We're going to need some analysis weights, as I mentioned, because we want to produce population estimates. We're going to use an aged interview variable, a gender variable. And then if we scroll down, you can see we have some here options for cognition. And I know that this total word recall score, you can see we have the immediate word recall score, delayed word recall score. But the total word recall score is considered a summary score. So we can search for that. You can see that we get these variables here, both from the RenHRS and the harmonized ELSA. You can see we have a couple of different weights here for the harmonized ELSA. You could click into any of these to find out more. So this is one way. The other way is the code books. And the code books have so much documentation. Let me pull up our... So this is the harmonized ELSA code book here. And you can see it's structured into different sections. So we have a section on demographics, identifiers and weights, health, insurance, cognition throughout here. We also have a contents. So for instance, we could go into cognition. If we wanted to know more about our cognitive summary score and inside of cognition, we could go to summary score. You're going to see this looks exactly like kind of the website. What we see here is our total recall summary score. So for wave five of ELSA, we're going to be using this variable for the respondent, which is r5 to r20, our word recall summary score. You can find here information about how it was constructed, any cross-wave differences. So if this kind of these questions change in different ways of ELSA, it's really important to mention that, because these are longitudinal surveys. And then there's any differences with the RenHRS. Of course, there were no differences here. All right. So now we know what variables we're going to use here. And then we also want to identify the relevant variables for the RenHRS, which we saw these using the concordant search on the gateway. But you can also use the RenHRS as a great code book, very similar to our code book. You could also search through to find these variables. One thing you'll note here is that we're using wave 10 of the RenHRS data for 2010, because the HRS started in 1992, whereas ELSA started in 2002. So our variable names are going to be a little different, because as you remember, our variable naming convention uses the wave, not the year. It's one of the things we're going to need to adjust for when we create a combined data set. So the first thing we're going to do is we're going to create a pool data set with variables from both harmonized data sets. So here's an example of this data code. So the first thing we would do is we would use, and we're just going to specify just these four variables, the variables we're going to be using from the RenHRS data. It's version O of the RenHRS data. And then we're going to append, and what append does in this data is it places the observations from the appending data set right below the current or using data set. So we're going to keep these variables. You'll see the other thing that we do here is we assign, we want to know, we don't want to mix up our observations. So we want to know who are those observations which are coming from the RenHRS, which represent the U.S., and which are those observations which come from ELSA, which represent England. So we're going to use country codes here, and these are ISO country codes, and I make a variable here called country, and I assign 840, the country code for the U.S., and the country code for England is 826. So I'm going to jump over to Stata, and I'll just show you this, and you'll have these commands also for yourself. So we have a couple of questions here. Someone asked, can we use version P of the RenHRS? Yes, absolutely. It would work absolutely fine. So we're going to use this release, and the software that I'm using is Stata, which is a statistical analysis package, which is similar to SAS or SPSS. So you could do this in either of those packages. All right. So we've got our harmonized ELSA data in there. That's perfect. We're going to add labels to identify our country identifiers. So if we do that inside of Stata, you could also tab that. You can see that we've got 37,000, a little more than 37,000 observations from the U.S. and almost 18,000 from England. This, of course, represents the U.S. studies a little larger, and that the U.S. study has been going on longer. So we currently have everyone now who's ever been in either of these studies currently in it. We're not going to use all these people, but it's fine for them to be loaded into Stata right now. So we also need to adjust for differing wave numbers. So as I mentioned, the variable names use wave numbers, and we're going to use year numbers instead. So we're going to use our, we're going to make a cognition variable, and we're going to call it R2,010 cognition. So we're going to say R2,010 cognition is equal to R10 TR20. If our country variable is equal to 840, meaning they're from the HRS data and represent the U.S., and we're going to replace that same variable with R5 TR20 if our country code is equal to 826, meaning these are people from ELSA and represent England. We'll do the same thing with our weight variable. You can see that we have two different weights, and we'll talk a little bit about weighting really shortly. And then we're also going to do the same thing with our age variable. You notice we didn't have to adjust our gender variable. The gender variable was R-A gender, and that's the same in both studies because it is not a wave-specific variable. It is considered a panel-specific variable, and that gender is something that doesn't change over time. So if we jump back into STATA, we can make these changes. So we adjust, we make our weight variable, our age variable, and our cognition variable. Sometimes it's helpful because we want to look over different age categories if you remember what we wanted to do here. So sometimes it's quite helpful to look at, not ages as a continuous variable, but different age categories. So for this analysis, we're going to look at these age categories here, and these are people who are 55 to 59, 60 through 64, 65 to 69, 70 through 74, 75 to 79, and then 80 and over. So we're going to represent these age categories here as kind of a way to look at how this cognitive summary score, this variable cognitive summary score, might be changing over different age groups. And so I'm using this eGen function. If you're not familiar, it's just kind of an enhanced variable generation function that allows you to do things like kind of cut a variable at the values. So we use that here. If we jump into this data, we can do that. And you can see here that if we tab this variable, so now here are our categories. So you can see we've got 55 and over, 60 and over, 65 and over, 70 and over, 75 and over, and 80 and over. So we've got our six age groups there. That looks perfect. You'll also notice there's a lot of people who don't fit this criteria. We have a missing value here for 28,000. And this is because both ELSA and HRS start from, also start tracking people at 50 and the HRS starts at 51. But both also include non-responder spouses which are not age-eligible to be in the survey. So there's actually a handful of quite young people in the survey because they are married to someone who is age-eligible, which is really helpful if you want to consider a family or a couple. You want both of those people's information. But in this case, for cognition, we're just looking at respondent level so we don't need to worry about those extra people. And we really want to see how cognition declines after age 55. We wouldn't expect a lot of change between 50 and 55 necessarily. So we've got our age category generated. So now let's talk about weighting. We're going to be using the SV set command. This is one of the reasons this data is very helpful is there's a lot of kind of built-in commands, particularly for people who are analyzing survey data and complex survey data. So we can say SV set. We can set our weight. This is a, we're saying it's a probability weight here. That's what P-weight means. And we're using the weight variable which has both the values from the weight variable from the HRS and from ELSA, which you might say these weight variables are probably generated differently and that's absolutely true. And we should take to account that. And we do that by setting our strata as our country variable. So we're saying these weight variables are really specific to the strata. And in this strata we have two different country variables. So if we jump back into this data, we can go through that and we can find a little more information. So you can see here that once we've set kind of our weighting, we have two strata. You can see the number of observations in both of those. And you can see our country codes here which is 826 and 840. The nice thing about using SVOI set is once this is set, we can call a lot of commands quite quickly without having to kind of redefine what is the weight and what is our strata. So the first thing we're going to do is we're going to estimate mean cognition just for each country. So for everyone kind of in the group that we're looking at, so 55 and over, what's the mean cognition in both countries? So here we say SVY. We need to specify a subpopulation because again we're not looking at everyone. Just people 55 and over. I'm just using 110 here kind of as a max age. And we're saying the mean cognition of our cognition summary score, which we're calling R 2010 cog, and we're saying over country. So if we jump back into this data we can go through. You can see that for England we would estimate for this age group of people who are 55 and over, we would estimate that people on average would recall 10 words in England and a little bit below 10 words in the U.S., 9.95 in the U.S. And we can also look at this graphically because sometimes that's a little easier and the graphs will help as we add more groups here. So for England you can see it's a little bit higher and it does look like it might be statistically significant a little bit higher. The ability to remember over 10 words and less than 10 words here. But not a huge difference. And you can imagine there's lots of reasons why this difference might occur. So when a lot of people use these data sets for cross-country analysis, instead of just saying there's a difference between the England and the U.S. we'll often kind of use what is called a difference in difference approach. So to say something like, well in the U.S. we see a difference in genders and this difference is different than the difference in genders that we see in the U.S. And it's a more interesting way that it doesn't require that we adjust for different things that could be happening in the U.S. and England and asking the same set of words. And the set of words are different in these two studies for instance. So that could adjust also. But if we're able to see that men always do better in England and women always do better in the U.S. then that difference is more interesting because we're comparing just people in England versus people in the U.S. and how those patterns might be different. They might mean cognition for each gender and each country. So we still need to set our subpopulation here 55 and over. And the only thing we have to change from our last command is that instead of this over and our over option here instead of just saying country we're going to say country and our gender. And what data knows to do there is to kind of cross tab those two variables and so we're going to get four estimates. One for males in England, one for females in the U.S. and one for females in the U.S. If we jump back into this data we can do this. You can see our estimate here for males in England is 9.95 words. It looks like it's quite higher for women here. It's 10.34 words for males in the U.S. It's 9.4. So again, significantly below both men and women in England and 10.33 meaning that females are actually quite high. It looks like they're really quite on par with females in England. If we look at this graphically we can see that this looks pretty much exactly true. As we read, so you do see a significant difference it looks like between men and women in both countries but that difference seems much more pronounced in the U.S. In particular, females in England are remembering we're estimating to remember the same number of words as females in the U.S. but men are significantly lower. U.S. men are significantly lower than both groups. So this is quite interesting. The last thing of course is this is a retirement and aging longevity study, both of these. So we want to look over age groups. One of the nice things about this verbal cognition summary score test is that it changes over age groups. So what if we use our age category variable and we can estimate cognition by country, gender, age groups. We're going to get even more categories here. So then we can again just use our SVY command so it recalls what we've already told this data about how our sample is designed and what the weight is. We're going to estimate mean cognition. We don't have to include our subpopulation here because in our over category we've added our age category variable already. So that's there. So that's already taken care of. If someone's not in that age category variable, they're not going to get an estimate. And we can also test this difference. As you'll see when we jump back into this data, you can test the difference and this is going to be an adjusted wall test for the difference in these estimates to make sure there's an estimate as we've been seeing in our graphs. So back in this data, you can see we get 24 subpopulations which makes sense. So they go from 55 to 59 males in England all the way through our 24th subpopulation which is females age 80 and over in the U.S. We get a bunch of estimates and if we go through we can test these subpopulations. So first if we test the significance of difference for our youngest age group in England. So this is the difference between men who are 55 to 59 and women who are 55 to 59. We see there is a statistically significant difference as we saw before. And then if we look at our oldest age group or our youngest age group in the U.S. again, so this is males and females in the same age group with 55 to 59 in the U.S. We see a significant difference. You will notice from estimates up here that this difference actually seems to be greater than it is for males and females in England at that age group. And this is one of the things that we saw in that graph that was kind of suggested on that graph before we looked at age. So let's look at our oldest age groups. So is there a significant difference in age groups in England? So English males and females who are 80 and over versus each other. And it looks like there is not a significant difference. So we are seeing something happen kind of on a gender perspective where we are not seeing the same pattern across all age groups. And if we look for the U.S. for the oldest age groups in the U.S. we do see this gender difference still come out. So let's look at this graphically because it's a little bit easier to estimate, to look at graphically. So here is our chart with all our kind of 24 groups here. You can see they are grouped so the genders are beside each other. So you can see again we do see this difference. We see a larger difference for our younger age groups between male and female, both in the U.S. and England. It's larger in the U.S. And you can see one of the things that you notice with both graphs is that it declines over ages in both countries. So that makes sense. Of course, the kind of interesting thing that we saw from our test there is that it looks like by our oldest age group in England there is not a significant gender difference. They look exactly similar. And the gender difference in the U.S. and cognition still looks just as strong as any other test. The other thing that you might notice is that it looks like the gradient is somewhat different for what we're estimating the number of average words that are recalled. So even though females in the U.S. and England in our youngest age groups started roughly the same number of words recalled by our oldest age group in the U.S., women are remembering it looks like maybe around nine words. But it looks like they're remembering maybe I would say something like seven or six and a half words in England. So it seems like the gradient for England is of course much steeper which is really interesting. And this is really just the first part of this puzzle. And really one of the powerful things about using these data sets is of course this is not just a survey about cognition. Cognition is one part of this test. So what are the things that lead to cognition? So both of these surveys have questions about education. For instance, what is our childhood education? What was the education different, was it gendered differently for our oldest age group here between England and the U.S. is something you could easily look at. These are longitudinal surveys also and we haven't really used that aspect here. So you might want to think of who's leaving the surveys. Who's not staying in this survey until their age 80 or over. And it could be that women of lower cognition are dropping out in the U.S. And that can be because of not follow-up or that can be because of end of life. And these people could be staying in England. And because these are longitudinal surveys, we could create what's called a balance sample. And we could follow people between say 2002 over the next all the way 2014. We could look over that age group between the U.S. and England and look at actual people who stay in these two surveys and compare how they are. That's a little bit longer analysis. But it's something you want to look at as we kind of get these differences. Of course there's also differences in employment. We know that how long people stay in employment often leads to cognition. You see a lot of cognitive declines often when people go out of the workforce. There's employment variables in all of these. And they also have a whole host of variables in both these surveys about how well people are integrated into their societies. How often do they see their friends, family, these communication networks having conversations either on phone or email that we also know can contribute to cognition, especially this verbal memory cognition. So this was just a really quick analysis to kind of give you an idea of some of the benefits of using harmonized data to be able to really quickly. I mean there's a lot of work that went into having these variables which are so easy for you to get started with your research. I will say really briefly that a harmonized ELSA and a ranked HRS do not contain every question which is asked in the ELSA and the HRS. They really focus on those variables which are one, asked across multiple studies, so can be compared to each other. And two are most often used for research. So as you saw in the codebook, the codebook for the harmonized ELSA is quite long. There's a lot of variables there, but it's not every variable in ELSA. But you can always merge in the original ELSA data with the harmonized ELSA data. If there's a specific variable that you want to bring in that isn't already included in the harmonized ELSA data. If you have any questions about using our site, the Gateway to Global Aging data, getting access to different harmonized data sets, questions about what's available, there's a lot of resources. I would encourage you to check out our website and just browse around. We also have the ability for you to generate some of your own graphs and tables from all of these studies. You can always email us at help at g2aging.org. We have a whole team of people who work on this here at the University of Southern California. And we can field your questions. Kind of the best person we always try to get back with in 24 hours.