1 00:00:00,025 --> 00:00:04,650 Hi, today we're going to be learning how to analyze our StreetSeen surveys. 2 00:00:04,650 --> 00:00:08,362 So it's been really great to see so many people developing so many different, 3 00:00:08,362 --> 00:00:12,317 interesting surveys. Asking about all kinds of interesting 4 00:00:12,317 --> 00:00:16,270 things from where do you think it's safer for our children to play to the busyness 5 00:00:16,270 --> 00:00:21,080 of a street. Which street would you like to walk down? 6 00:00:21,080 --> 00:00:25,550 which park would you prefer to play in? Lots of different questions about our 7 00:00:25,550 --> 00:00:28,282 cities. So, I'm going to walk you through step by 8 00:00:28,282 --> 00:00:32,297 step how to analyze StreetSeen. And I'll give you one option. 9 00:00:32,297 --> 00:00:36,617 Which is the straight forward simple way to analyze StreetSeen and draw some basic 10 00:00:36,617 --> 00:00:40,081 conclusions. And then I'm going to explain how to, 11 00:00:40,081 --> 00:00:44,930 conduct a more detailed discrete choice model way of analyzing this. 12 00:00:44,930 --> 00:00:47,270 And this requires a statistical background. 13 00:00:47,270 --> 00:00:50,510 So if you know how to use statistics, then I'm going to be showing you some 14 00:00:50,510 --> 00:00:55,460 ways that you can dig deeper and create statistically significant results. 15 00:00:55,460 --> 00:00:58,725 Many of you participated in this simple survey. 16 00:00:58,725 --> 00:01:02,611 Bicycling preferences in Columbus Ohio, so we wanted to know which street would 17 00:01:02,611 --> 00:01:07,785 you prefer to ride a bicycle down 1 or 2? And so one of the key things in designing 18 00:01:07,785 --> 00:01:11,384 a StreetSeen survey is that you want to consider what is it that your asking 19 00:01:11,384 --> 00:01:15,306 people about. So in this case which street would you 20 00:01:15,306 --> 00:01:18,612 prefer to ride a bicycle on and then, we want to understand what are the 21 00:01:18,612 --> 00:01:23,810 differences between these two images. And amongst the images that you're 22 00:01:23,810 --> 00:01:27,612 particularly interested in. So a discrete choice model is a method of 23 00:01:27,612 --> 00:01:31,836 describing, explaining, and predicting choices between two or more discrete 24 00:01:31,836 --> 00:01:35,784 alternatives. So for example, in the picture that you 25 00:01:35,784 --> 00:01:39,546 saw before we had two discrete choices that you can choose between, the picture 26 00:01:39,546 --> 00:01:44,997 on the right or the picture on the left. And what we want to be able to do is just 27 00:01:44,997 --> 00:01:48,702 statistically relate the choices made by each person to the attributes of the 28 00:01:48,702 --> 00:01:54,270 alternatives available to the person. So in each case, you were given two 29 00:01:54,270 --> 00:01:57,850 choices and you had to make a choice between each one of those. 30 00:01:57,850 --> 00:02:02,074 And so we want to know which people chose which images and what are attributes of 31 00:02:02,074 --> 00:02:06,225 those alternatives. So what's the difference between image 32 00:02:06,225 --> 00:02:09,360 one and image two? And when we're doing a discrete choice 33 00:02:09,360 --> 00:02:12,520 model we want the set of alternatives to be exhaustive. 34 00:02:12,520 --> 00:02:15,496 And we want it to be mutually exclusive and we want it to be a finite number of 35 00:02:15,496 --> 00:02:19,335 alternatives. So, for example, you could only choose 36 00:02:19,335 --> 00:02:23,484 one of the two images. And that the alternatives, there were 37 00:02:23,484 --> 00:02:27,114 finite set of alternatives. So, for example, you selected a number, a 38 00:02:27,114 --> 00:02:30,562 fixed number of images. In my case, there were 60 different 39 00:02:30,562 --> 00:02:33,948 images in that image set. And so, you were voting between two 40 00:02:33,948 --> 00:02:38,310 images, and there were only a certain number of images that were in the set. 41 00:02:38,310 --> 00:02:41,090 Although, you could vote infinitely if you wanted to. 42 00:02:41,090 --> 00:02:45,474 You were limited to those 60 images that were randomly shown in pairs. 43 00:02:45,474 --> 00:02:51,110 Now in this case, we want a defined set of variables. 44 00:02:51,110 --> 00:02:54,892 So when you look at the images that you've selected, you've identified key 45 00:02:54,892 --> 00:02:59,388 characteristics, okay? So in my example I selected ten variables 46 00:02:59,388 --> 00:03:03,700 that were different between the various images. 47 00:03:03,700 --> 00:03:07,110 So the amount of traffic. So there might be zero cars shown in a 48 00:03:07,110 --> 00:03:12,050 particular image, up to ten or more cars. And so I just classify in different 49 00:03:12,050 --> 00:03:16,840 segment the number of vehicles that were seen on the street. 50 00:03:16,840 --> 00:03:19,990 Whether or not there was a sidewalk, whether there was a sidewalk on one side 51 00:03:19,990 --> 00:03:23,050 or whether there was a sidewalk on both sides. 52 00:03:23,050 --> 00:03:26,440 I looked at the type of parking. Was there no parking on the street? 53 00:03:26,440 --> 00:03:30,440 Parallel parking, pull-in parking or a parking lot. 54 00:03:30,440 --> 00:03:32,600 I looked at the character of the street surface. 55 00:03:32,600 --> 00:03:37,020 So was it a well maintained street? Was it a poorly maintained street? 56 00:03:37,020 --> 00:03:39,480 Was it a brick street? An asphalt street? 57 00:03:39,480 --> 00:03:42,530 A concrete street? I looked at the number of lanes. 58 00:03:42,530 --> 00:03:46,290 Is it a simple one lane alley? Is it a two lane street? 59 00:03:46,290 --> 00:03:50,529 A six or more lane street? And I characterized it, each image by the 60 00:03:50,529 --> 00:03:53,540 number of lanes. I identified whether or not any 61 00:03:53,540 --> 00:03:56,840 pedestrians could be seen in the image, or whether any bicyclists could be seen 62 00:03:56,840 --> 00:04:00,466 in the image. I look at the grade of the street so, for 63 00:04:00,466 --> 00:04:03,900 example, is it a flat street? Is it a curve street? 64 00:04:03,900 --> 00:04:07,352 Is it a hilly street? And then I looked at the different kinds 65 00:04:07,352 --> 00:04:11,008 of land uses. So is this a street where there are 66 00:04:11,008 --> 00:04:15,170 individual houses? Are there apartments? 67 00:04:15,170 --> 00:04:18,580 Or, is it a downtown with office buildings, or what are, what is the 68 00:04:18,580 --> 00:04:22,862 character of the land use? I looked at trees, so are the trees along 69 00:04:22,862 --> 00:04:26,050 the street? Are they set back from the street? 70 00:04:26,050 --> 00:04:29,949 Or, is it a heavily forested area? And then I looked at street calming 71 00:04:29,949 --> 00:04:33,126 measures. So for example were there pedestrian 72 00:04:33,126 --> 00:04:36,720 crossings? Were there bumps in the road to calm 73 00:04:36,720 --> 00:04:39,822 traffic? And so these were the ten variables that 74 00:04:39,822 --> 00:04:42,395 I looked at. In your case, you'll have your own set of 75 00:04:42,395 --> 00:04:46,089 variables that you were thinking about. So when you, if you're thinking about 76 00:04:46,089 --> 00:04:49,470 parks, or is this safe for children, what are the things that are different between 77 00:04:49,470 --> 00:04:53,380 the images? And so you may not have thought about 78 00:04:53,380 --> 00:04:56,667 this in this type of detail when you created the survey. 79 00:04:56,667 --> 00:04:59,574 But you can just as easily go back now, and think through, what are the 80 00:04:59,574 --> 00:05:02,560 differences between these different images? 81 00:05:02,560 --> 00:05:07,569 In my own case, I looked at literature around streets and bicycling. 82 00:05:07,569 --> 00:05:11,664 To better understand what are the characteristics that are in the research 83 00:05:11,664 --> 00:05:15,580 that identify. What makes people want to bicycle or not 84 00:05:15,580 --> 00:05:19,630 bicycle on a particular street. No, no, I'm not asking you to go into 85 00:05:19,630 --> 00:05:22,614 that level of detail for this particular exercise. 86 00:05:22,614 --> 00:05:25,664 But it's something you can keep in mind, and certainly you can use your own 87 00:05:25,664 --> 00:05:29,506 intuition about what are the differences between the images. 88 00:05:29,506 --> 00:05:31,675 And how do you understand the differences. 89 00:05:31,675 --> 00:05:36,640 Now, you get your results. So when you go to StreetSeen, you can 90 00:05:36,640 --> 00:05:39,430 simply click on the Analyze button for your survey. 91 00:05:39,430 --> 00:05:43,780 And it's going to pull up these results. And so you'll see, in this case, that 92 00:05:43,780 --> 00:05:48,400 this partic, these series of images were favored 75% of the time with the top 93 00:05:48,400 --> 00:05:54,380 image and it goes down from there. Now, I'm just going to to take two 94 00:05:54,380 --> 00:06:00,590 images, the image that was most preferred and the image that was least preferred. 95 00:06:00,590 --> 00:06:03,824 And so when you look at these two images, what are the things that you notice that 96 00:06:03,824 --> 00:06:06,839 are different? And so we can think back to the variables 97 00:06:06,839 --> 00:06:10,838 that we were referencing. So in this case, the top image, there's 98 00:06:10,838 --> 00:06:13,677 no traffic. And in the bottom image, there's a lot of 99 00:06:13,677 --> 00:06:16,978 traffic. In the top image, this is a two-lane 100 00:06:16,978 --> 00:06:20,502 street. In the bottom image this looks like it's 101 00:06:20,502 --> 00:06:25,496 a four-plus-lane street. When we look at the top image, there's a 102 00:06:25,496 --> 00:06:28,998 lot of landscaping. That there are streets along the tree the 103 00:06:28,998 --> 00:06:32,510 road, as well as trees that are set back from the road. 104 00:06:32,510 --> 00:06:37,720 In the bottom image yes, there are trees. They're set back in the distance. 105 00:06:37,720 --> 00:06:41,230 we can see that in this case there are sidewalks on both side of the road, and 106 00:06:41,230 --> 00:06:45,770 in the bottom case there are no sidewalks that are visible, right? 107 00:06:45,770 --> 00:06:49,460 So you can continue to go through this process of looking at each variable. 108 00:06:49,460 --> 00:06:52,882 And so for the simple analysis, I would ask for you to look at the top five 109 00:06:52,882 --> 00:06:58,260 images, and the bottom five images. And understand what are the differences. 110 00:06:58,260 --> 00:07:02,068 So using your ability to simply measure, what's different about these different 111 00:07:02,068 --> 00:07:06,670 images, and how do they compare? So why are people voting for certain 112 00:07:06,670 --> 00:07:10,576 images over other images? So this is just a really simple analysis 113 00:07:10,576 --> 00:07:13,904 for you to be able to explain what the differences were between these various 114 00:07:13,904 --> 00:07:17,163 images. Some of you may have wanted to compare 115 00:07:17,163 --> 00:07:19,857 different areas. So, you might've compared different 116 00:07:19,857 --> 00:07:22,840 cities, or you might've compared different neighborhoods. 117 00:07:22,840 --> 00:07:27,460 And, so you'll want to report out which areas were favored more than others. 118 00:07:27,460 --> 00:07:30,160 In my own case, this is irrelevant for me. 119 00:07:30,160 --> 00:07:33,210 I did wind up picking some additional areas of the city, but it's because I 120 00:07:33,210 --> 00:07:36,820 wanted to get different types of streets to be included. 121 00:07:36,820 --> 00:07:41,167 But all of them for from within Columbus, and I don't see that it's particularly 122 00:07:41,167 --> 00:07:46,662 valid to compare between areas, and the percent favored is quite small. 123 00:07:46,662 --> 00:07:50,442 Then we have the heat map, for most of you the heat map is going to be 124 00:07:50,442 --> 00:07:54,520 irrelevant. This really depends on having a very 125 00:07:54,520 --> 00:07:58,810 large number of votes that are included in your study. 126 00:07:58,810 --> 00:08:03,088 So if you have an extremely high number of votes, then colors of the map will 127 00:08:03,088 --> 00:08:08,413 change and what will be shown. Is that where there are areas that are 128 00:08:08,413 --> 00:08:12,702 most favored then the heat maps are going to grow. 129 00:08:12,702 --> 00:08:16,980 And you'll see higher concentrations of the heat that is shown on these 130 00:08:16,980 --> 00:08:20,626 individual maps. But for most of you, you have a small 131 00:08:20,626 --> 00:08:25,339 number of votes and so it's just going to show each individual location as a dot. 132 00:08:25,339 --> 00:08:27,720 And that's not going to be very helpful to you. 133 00:08:27,720 --> 00:08:30,575 So, for most of you, you don't need to worry about this heat map. 134 00:08:30,575 --> 00:08:34,481 Okay, now those of you who want an extra challenge, you know something about 135 00:08:34,481 --> 00:08:38,697 statistics you want to take on and really understand things in a more significant 136 00:08:38,697 --> 00:08:42,720 way. I'm going to walk you through how to do a 137 00:08:42,720 --> 00:08:46,162 discrete choice model. The first thing you're going to do is 138 00:08:46,162 --> 00:08:50,284 you're going to go the results and you're going to download the full results. 139 00:08:50,284 --> 00:08:53,602 You can download it as a CSV or an XLS file, it's really up to you. 140 00:08:53,602 --> 00:08:57,082 Now, I don't have time in this class to teach you how to use statistical 141 00:08:57,082 --> 00:09:00,544 programs. But I am going to offer you some simple 142 00:09:00,544 --> 00:09:03,291 instructions. So first of all, you can use a 143 00:09:03,291 --> 00:09:07,281 statistical program, if you already have one, you're welcome to use whatever you'd 144 00:09:07,281 --> 00:09:10,927 like. Or you can use a free statistical program 145 00:09:10,927 --> 00:09:17,153 and the one I'm going to recommend is R. R is available at r-project.org's 146 00:09:17,153 --> 00:09:20,610 website. And then one of my faculty colleagues 147 00:09:20,610 --> 00:09:24,090 Professor Brighton at Ohio State University has created a How to Use a 148 00:09:24,090 --> 00:09:28,925 Discrete Choice Model. And so I'll be providing that link up on 149 00:09:28,925 --> 00:09:32,696 the website as well. And he walks you step by step, how to run 150 00:09:32,696 --> 00:09:37,776 a discrete choice model using R, okay? So this is a free, simple way, and if you 151 00:09:37,776 --> 00:09:41,440 want this challenge you can walk through it and see how it works. 152 00:09:41,440 --> 00:09:44,740 I ran the discrete model, and I got the results. 153 00:09:44,740 --> 00:09:49,220 And so in a very simple explanation, the good news is that the model results match 154 00:09:49,220 --> 00:09:53,890 my hypotheses about what I thought was going to happen. 155 00:09:53,890 --> 00:09:57,245 So for example, I believed that the larger the number of vehicles that are 156 00:09:57,245 --> 00:10:00,390 seen. The less likely people would be to choose 157 00:10:00,390 --> 00:10:03,880 that particular image when paired with another image. 158 00:10:03,880 --> 00:10:07,840 And that's in fact what I found, is that the larger the number of vehicles and the 159 00:10:07,840 --> 00:10:13,400 larger the number of lanes, the less likely an image is to be selected. 160 00:10:13,400 --> 00:10:17,333 On the positive side, where you have sidewalks, where you have pedestrians and 161 00:10:17,333 --> 00:10:22,430 bicyclists and where you have, trees along the, the side of the road. 162 00:10:22,430 --> 00:10:25,768 Then you are more likely to choose that particular image. 163 00:10:25,768 --> 00:10:31,120 And what was interesting to me is that I expected that parking would be negative. 164 00:10:31,120 --> 00:10:34,081 That if you had parallel parking, that you would be less likely to choose the 165 00:10:34,081 --> 00:10:36,839 image. Because there's a danger to bicyclists 166 00:10:36,839 --> 00:10:41,220 from opening the door, and so you can run into the car door and be injured. 167 00:10:41,220 --> 00:10:44,433 Or if it's pull-in parking, then people can't see you as easily when they're 168 00:10:44,433 --> 00:10:48,914 trying to back out of that parking space. Or, in a parking lot that you could have 169 00:10:48,914 --> 00:10:51,700 cars that are pulling out from a parking lot. 170 00:10:51,700 --> 00:10:55,670 But in my model this was not statistically significant. 171 00:10:55,670 --> 00:10:57,570 Alright?, so, that was an interesting finding for me. 172 00:10:57,570 --> 00:11:00,298 So, those were the variables that I was looking at and I was happy to see that 173 00:11:00,298 --> 00:11:03,540 the model matched the results that I, I thought about. 174 00:11:03,540 --> 00:11:08,326 Now, the next step. Is, well, are there differences between 175 00:11:08,326 --> 00:11:11,630 people who are in Columbus or another area? 176 00:11:11,630 --> 00:11:15,050 Are there differences between the different countries that people are from? 177 00:11:15,050 --> 00:11:19,137 And one of the things that's captured by StreetSeen is the voter latitude and the 178 00:11:19,137 --> 00:11:22,907 voter longitude. And so what you'll be able to do is use 179 00:11:22,907 --> 00:11:26,994 the voter latitude and longitude to determine the location of an individual 180 00:11:26,994 --> 00:11:32,700 that took your survey. So, you can go to any number of free 181 00:11:32,700 --> 00:11:36,200 websites that provide the latitude and longitude. 182 00:11:36,200 --> 00:11:40,507 So what you can do at latlong.net, is you can click the LatLong to Address button, 183 00:11:40,507 --> 00:11:44,704 and then you can enter that latitude and longitude. 184 00:11:44,704 --> 00:11:47,586 And then you can find out where people are from. 185 00:11:47,586 --> 00:11:50,536 From and you can enter that country or city or whatever it is that you're 186 00:11:50,536 --> 00:11:54,720 particularly interested in. So in this example this is the latitude 187 00:11:54,720 --> 00:11:59,354 and longitude for the Taj Mahal. And so you can go in and as you saw in 188 00:11:59,354 --> 00:12:04,840 this list, I have created the gone in and filled in the latitude and longitude. 189 00:12:04,840 --> 00:12:08,497 And I can see that somebody completed the survey from Buenos Aires in Argentina 190 00:12:08,497 --> 00:12:13,408 from a number of locations in Australia. In my particular case, what I'm really 191 00:12:13,408 --> 00:12:16,970 interested in is comparing the results for Columbus. 192 00:12:16,970 --> 00:12:20,392 So, for example, many of the people who took my survey may recognize the 193 00:12:20,392 --> 00:12:24,072 individual streets. And so may have different opinions 194 00:12:24,072 --> 00:12:28,340 because they have a user experience from bicycling on the streets. 195 00:12:28,340 --> 00:12:32,010 And so I want to understand whether or not what they've said is different. 196 00:12:32,010 --> 00:12:34,290 From what the people who are not in Columbus have said. 197 00:12:34,290 --> 00:12:37,700 And then I want to understand some of the cultural differences so are there 198 00:12:37,700 --> 00:12:41,433 differences between what people in Asia might think. 199 00:12:41,433 --> 00:12:44,843 From what people in South America might think and have preferences around 200 00:12:44,843 --> 00:12:48,150 bicycling. So that's the next step that I'll be 201 00:12:48,150 --> 00:12:51,500 undertaking is integrating that into my discrete choice model to understand if 202 00:12:51,500 --> 00:12:55,559 there are differences. And this is fairly simple to be able to 203 00:12:55,559 --> 00:12:58,170 set up. I can fragment my samples and you're 204 00:12:58,170 --> 00:13:01,210 welcome to do this in your own studies as well. 205 00:13:01,210 --> 00:13:06,040 so I encourage you to experiment, to play with the data and to just have some fun. 206 00:13:06,040 --> 00:13:09,800 Seeing how you can use discrete choice models in a visual survey setting. 207 00:13:09,800 --> 00:13:12,874 And the last thing I would say is if you're not a statistical person, or 208 00:13:12,874 --> 00:13:16,478 somebody that has these statistical skills, you're welcome to try them using 209 00:13:16,478 --> 00:13:21,130 the R and the walkthrough, but if you're not that's okay. 210 00:13:21,130 --> 00:13:24,330 Stick with some of the simple analysis that I've showed you how to use, by just 211 00:13:24,330 --> 00:13:28,610 understanding and comparing the the images that are in your data set. 212 00:13:28,610 --> 00:13:30,890 Good luck. I look forward to seeing your results.