Carrie

Farmers Markets of North Carolina

Carrie Weiner Campbell, Leondra Edwards, Pam Lach & Sarah Dooley

Project Proposal

Topic

We propose a visualization of farmers markets in North Carolina. We will use farmers market data available for download from http://www.data.gov/tools/4034 and http://apps.ams.usda.gov/FarmersMarkets/ . We have already downloaded the basic farmers market data (excel file -- attached below), which is nominal and spatial (farmers market names by state with associated latitude and longitude coordinates). Since the original data set is rather large, we have decided to focus on North Carolina to make the visualization more relevant to us.

To this data, we will add a series of attributes, including:

• • product availability (temporal data -- by season or month, see What's In Season)

  • payment method (cash, credit, etc)

  • participation in federal nutrition programs (*Supplemental Nutrition Assistance Program (SNAP, formerly known as food stamps), Women, Infant and Children (WIC) , WIC Cash Value Vouchers (WIC CVV) and Senior Farmers Market Nutrition Programs (SFMNP))

  • possibly median household incomes per county and/or health data (obesity rates per county, etc)

Why This Interests Us

As a group of people who rely on farmers markets in our own lives, we are interested in the range of farmers markets available to folks across North Carolina, and in particular to look for correlations between time, space, and economics (access to fresh produce). For instance, is there a correlation between income level and access to produce? Ultimately, we envision this visualization as a tool for folks who are looking to locate farmers markets in their area, and what methods of payment they can use (since that could be a barrier for some).

Scope

To make this project more manageable, we will focus on North Carolina's farmers markets. Our scope and design considerations will be driven by a set of user tasks, such as

• • locate where/when to buy certain items with certain payment methods

  • Identify produce availability across the state (this is spatial and temporal)

  • Draw comparisons between different counties and/or regions across the state to identify under-served populations/areas

This will be a design for a map-based visualization with interactive capabilities for viewing the attributes listed above.

Assignment of Duties

We anticipate the following tasks will need to be accomplished:

• 1. continue gathering the data

  2. determine user tasks and levels of interaction

  3. determine visualization strategies: develop a plan for processing the data

  4. sketch out images/mockups/wireframes

  5. produce strategy document

We will each contribute to the first three tasks and will divide up equally among us the sketches. We will work on the strategy document as a team.

Expected Deliverable

For the midterm, we anticipate producing a series of wireframes and/or mockups featuring different map views. We have begun investigating several software options for producing our markups, including Mockingbird and Balsamiq. Balsamiq offers a free trial, while Mockingbird charges $9/month, which lets you create 2 user accounts that we can share.

We also plan to produce a strategy document where we outline the data sets we are using, how we will transform them, and how our visualizations will support our chosen user tasks.

Updated Project Specs(Feb 29)

Use Case: Allow users to look at map to determine where they can get certain produce at certain times of the year given particular payment options ("market basket" approach) just for NC (100 counties). Sort on attributes: date (by season), payment, crop, location.

Data Processing: We will need to hand-code data from http://search.ams.usda.gov/farmersmarkets/ for each county. Add to existing spreadsheet on farmers markets (attached here) -- payment type, produce available and/or product type availability

Division of Labor:

Data to acquire/Data to code:

• • payment methods for each farmers market: add to existing location spreadsheet which has been uploaded to Google Docs (so that we can work simultaneously without having to worry about version control)

  • each of us takes 25 counties (about 54 markets per person)

  • still looking for data about crop availability by county in NC (if we can't get it, we'll use the pre-existing product categories from USDA)

Though we are not doing a full-on visualization at this point, we still want to acquire the data to dump into Tableau or similar for initial views of the data, and for use in a final project.

Timeline for acquiring/coding the data: Wednesday @ noon.

Division of Labor for Final Deliverable: (to be added once we get some confirmation from Dr. H about whether the workload is suitable to the size of our group)

• 1. scenarios/use cases/interaction plans

  2. write up overall strategy/documentation (CUT-DDV discussion)

  3. static snapshots/designs

Tools to Use:

We're deciding between Tableau and ManyEyes (if we do interactive visualization). If we just snapshots we would use other tools.

Deliverable:

Still not sure if we need an Interactive visualization (group to email Dr. H to verify that). Else we'll snapshot the different filtered displays. Sample interactions plotted out - sample views.

Final Document: CUT-DDV discussion with sample interactions/views.

For now we are not going to handle other county data (rates of obesity, poverty, unemployment, etc). We have found these data in several places (NC Atlas and NC Rural Center) but decided to hold off on this. If we decide to continue this project for the final we will add those data sets then.

Midterm Project Presentation

Background / data

For this project we were interested in exploring the presence of Farmers Markets across North Carolina, and how the presence or absence of markets, as well as the accessibility of those markets (location and financial) might be visualized. Adding to that, we decided to layer these data with information about NC counties, including poverty rates, obesity rates, presence of family-owned farms, etc.

For our visualizations, we pulled data from four main sources:

• • USDA farmers market data, including locations, product categories, and payments accepted (source)

  • County profiles from NC Rural Center Databank (source)

  • USDA food atlas data (source)

  • USDA food desert data (source)

Each of these sets contained many types of data. In particular, each used geography as an organizing feature (latitude / longitude, counties, or census tracts), which made mapping a logical choice for many of our visualizations. Continuous numerical data was also strongly represented, along with categorical data. Specific data treatments will be discussed for each visualization below.

All of these data sets needed to be cleaned up (in many cases, a county and/or state field had to be added) and much of the data had to be hand coded because it was not available for download. Each of us contributed to the data processing, with Leondra taking the lead in the farmers market data, Pam handling the County Profiles, Carrie managing the food atlas data, and Sarah dealing with the Food Desert data. We used Google Docs for our data to enable sharing and collating.

Task 1:Farmers Market Data, Locations, Products for Sale, Payment options

Leondra - Farmers Market Data

The goal was to answer three specific questions by using filters and showing a visualization to answer each of these questions:

• • Is there a Farmers Market in a particular city? (I allowed a filter to select a city to find a farmers market)

  • Will this farmers market take Credit, WIC, SFNMP?

  • What products are sold at the farmers market in the selected city?

My goal was to keep the visualization as simple as possible while providing as much information as the user needed to answer the above questions:

Data Used: Farmers Market Data Spreadsheet; this sheet contained all products sold, the type of payment taken, the name of the farmers market, street name, county, city

Market Locations Lat/Long

Software Used: Tableau 7.0

Included individual map and spreadsheet then added an additional dashboard for both. Within Tableau these are interactive spreadsheets and contains a mouse over to give more detail about each farmers market.

Screen shot of Map data along with filters:

This will answer the questions of Is there a Farmers Market in a particular city, Which payments does this farmers market take, Also it uses color coded to list if you can purchase prepared meals at this farmers market.

Sometimes a spreadsheet view a simple yet powerful way to list information. I also incorporated a spreadsheet view to answer specific questions. This one uses color coded to answer the question as to

if the farmers market takes credit as payment. Along with some options of things for sale, BakedGoods, Cheese

I also provided the option of the user being able to see both options at the same time by using dashboard to display both options.

Then I combined the aspects to see the interplay of merchandise variety and payment options. We were unable to do this for all 196 markets, because not all had provided details about their merchandise. For the bar chart above, we had data on 65 markets; those markets are also pictured in the treemap below, which I made with ManyEyes. A treemap, sorted by county, seemed the best way to answer the question, "Which market(s) within each county give people using SNAP/WIC credits the most variety?" The treemap can be accessed here. For purposes of the brief class presentation, here is a screenshot of it:

The size of each box indicates its amount of variety (larger box means more merchandise categories); the color indicates the number of payment options (darker boxes mean more options are available).

(To develop this idea further, as part of my final project, I want to create an interactive visualization that includes information about seasonal availability of crops. The useful website LocalHarvest.org includes detailed lists of what's on offer at dozens of North Carolina farmers markets. The website allowed for only manual extraction of this data, so given the allotted time, I took only a sample of the information -- for 22 markets -- and attempted to combine that crop info with payment options to see what a person using food stamps would be able to buy at various times of the year. However, this dataset, with its four dimensions (season: categorical or temporal; crop: categorical; payment types: 5 possible categories; and market: geospatial) proved to be too complex for my Tableau skills at this point. Undaunted, I aim to take up this challenge when I create my final class project.)

Task 3: Farmers markets and health and income factors

(Pam)

I was most interested in exploring connections and possible correlations between the presence or absence of farmers markets in NC with factors related to health (obesity and diabetes) as well as wealth. For instance, would there be a higher adult obesity rate in areas without farmers markets? Or are there income barriers to accessing fresh produce? I thought this would be an important visualization for researchers and policy makers looking to combat obesity and/or for folks looking to expand access to quality food for low-income populations (very important in this economy!).

I was most interested in exploring the spatial component of these relationships, and used Tableau to create a set of maps to tell several different stories. I accomplished this by merging the data from NC Rural Center's County Profiles (which were downloaded individually and merged into a single spreadsheet) with the USDA Food Atlas Data.

I attempted to merge this data with the Food Deserts in GeoCommons, but the geocoding didn't work due to the format of the data. When I tried uploading it into GeoCommons it failed miserably due to different formats: my data had separate County and State attributes (the state is required or else many of the counties get placed in the wrong location) but the Food Desert, based on Census Tracts, had the states coded differently.

One of the challenges here was to balance the number of attributes I wanted to show. There were only so many dimensions I could add to the actual map -- one attribute for size of bubble, another for color. So I decided to rely on the Quick Filters feature in Tableau. At first, I had everything in a single map, but quickly realized there was a point of diminishing returns. So I decided to split my maps into different categories: Health (obesity/diabetes), Poverty, Food Assistance, and Children (as a particularly at risk population in many cases). To facilitate comparisons, I used the number of farmers markets/county (2010) as the size of the bubble on each map. I applied the same scale to this variable for each. Ultimately, I went with an interactive set of small multiple maps in a Trellis format using Tableau and published with Tableau Public.

I found that the screen was too busy with all of my quick filters applied, as this screenshot of my original attempt shows:

To address the cluttered screen real estate, I removed all the filters from the small multiples version to provide a broad overview of the data. You can see this at: overview maps. Here's a screenshot:

This view limits interactivity for the user since there are no filters, but it makes for a cleaner first impression. To support zoom, filter, and details on demand, here are each of the four individual maps:

Map 1. Farmers Markets and Health Factors

• • size of bubble: number of farmers markets (2010)

  • color of bubble: adult obesity rate (2008)

  • quick filters: adult diabetes rate (2008), low-income preschool obesity rate (2009)

Screenshot:

Filtered View:

Link to interactive map: http://public.tableausoftware.com/views/CountyDataFarmAtlasFarmersMarkets/HealthFactors?:embed=y

Map 2. Farmers Markets and Poverty

• • size of bubble: number of farmers markets (2010)

  • color of bubble: per capita income (2005-2009)

  • quick filters: median household income (2010), number of unemployed (2010), unemployment rate (2010), poverty rate (2010)

Screenshot:

Link to interactive map: http://public.tableausoftware.com/views/CountyDataFarmAtlasFarmersMarkets/Poverty?:embed=y

Map 3. Farmers Markets and Food Assistance Programs

• • size of bubble: number of farmers markets (2010)

  • color of bubble: percent receiving food assistance (2010)

  • quick filters: total SNAP benefits $1,000 (2008)

Screenshot:

Link to interactive map: http://public.tableausoftware.com/views/CountyDataFarmAtlasFarmersMarkets/FoodAssistance?:embed=y

Map 4. Farmers Markets and Children/Risk Factors for Children

• • size of bubble: number of farmers markets (2010)

  • color of bubble: child poverty rate (2005-2009)

  • quick filters: Farm to School Program (2009), low-income preschool obesity rate (2009), per student expenditures K-12 (2010)

Screenshot:

Link to interactive map: http://public.tableausoftware.com/views/CountyDataFarmAtlasFarmersMarkets/Children?:embed=y

Another challenge is the with the data itself. Rates are not necessarily comparable across attributes, particularly with respect to time. Yet there wasn't enough temporal consistency to chart change over time. We had a lot of categorical data, which didn't seem to work in my maps, so I stuck with continuous data types (though some are rates, some are actual numbers, and some are percentages). The one exception was the binary data of Farm to School (Y/N).

Task 4: Farmers markets and agricultural data

(Sarah)

This set of visualizations addresses the task of determining relationships between farmers market locations and other county-based data about agriculture and food availability. I started by trying out a Google map for one of the relationships (here). On the upside, I liked the ability it provided to pan and zoom for more detail (which users would need to do to examine the more densely market-populated areas), and fusion tables were a great tool for mapping the data; on the downside, I ran into some limitations with styling (literally, it limits how many styles you can apply) and options for interactivity.

I ended up using Geocommons for my maps instead. It offered several advantages: multiple ways to geocode (good, since we had multiple kinds of geographic data); more and better styling options, including attractive and uncluttered base maps; the ability for viewers to toggle layers on and off; the ability to add multiple layers without problems. Overall, I think this is a great (free!) tool, and I'm fairly pleased with how things came out. To get my complaints out of the way up front, though, there were a few aspects that weren't ideal:

• • Although there are good styling options, they're not totally comprehensive—no textures, for instance, and no 3D. This tool therefore worked best for maps that only include 2-3 layers of data.

  • Control over the data processing is limited; for instance, it only offers a few set binning options, and none of them let you set the scale (hence some weird scales).

  • There doesn't seem to be any way to style the legend, and it is inappropriately gigantic.

  • The data can be slow to load, which is why I'm including screenshots here.

Quibbles aside, here are the maps I thought worked best given the strengths of this tool and the highlights of the data. These are intended to provide an at-a-glance view of patterns, with the opportunity to explore in more detail by panning and zooming. Click any image to go to the live map version.

Farmers market locations vs total farms (per county)

Data used:

-market locations (geographic; lat/long)

-farms per county (continuous in data set; binned into intervals for map)

Farmers market locations vs average farm size (per county)

Data used:

-market locations (geographic; lat/long)

-average farm size per county (continuous in data set; binned into intervals for map)

Farmers market locations vs % family-owned farms (per county)

Data used:

-market locations (geographic; lat/long)

-% family-owned farms per county (continuous in data set; binned into intervals for map)

Number of farms, average farm size, % family-owned farms (per county)

To provide added perspective on the trends shown in the maps above, I created another view that relates the three agricultural data points:

Data used:

-farms per county (continuous in data set; binned into intervals for map)

-average farm size per county (continuous in data set; binned into intervals for map)

-% family-owned farms per county (continuous in data set; binned into intervals for map)

Farmers market locations vs total agricultural receipts, crop receipts, & livestock receipts (per county)

Data used:

-market locations (geographic; lat/long)

-total agricultural receipts per county (continuous in data set; binned into intervals for map)

-total crop receipts per county (continuous in data set; binned into intervals for map)

-total livestock receipts per county (continuous in data set; binned into intervals for map)

Farmers market locations vs food deserts (by census tract)

Data used:

-market locations (geographic; lat/long)

-food deserts, as defined by the USDA (definition), represented here by % of population with low access to grocery stores (continuous in data set; binned into intervals for map)

Farmers market locations, Farm to School programs, & number of farms (per county)

Data used:

-market locations (geographic; lat/long)

-county participation in Farm to School program (yes / no)

-farms per county (continuous in data set; binned into intervals for map)

Data correlations chart

To provide a different view of the relationships examined in the maps, I also charted the correlations among different data points included. Interestingly, most of the correlations turned out to be pretty weak (with the exception of some of the economic factors). Compared to the maps, this approach better facilitates comparing the strengths of different relationships. On the other hand, the maps provide a sense of how what patterns there are, are geographic (often with the eastern or western parts of the state contrasting with the rest). I thought the lack of meaningful correlation shown here between Farm to School participation and factors that could point to the appropriateness of such a program suggested interesting potential for change—however, I've now realized that the data on that point was incomplete. Better data is available here should we want to incorporate this info more into an end-of-term project...

Data used are all continuous, except for Farm to School participation, which is binary.

Reviews

Review 1: Tali Beesley on Leondra's portion

This review is specifically for Leondra’s portion of the midterm. First, I appreciate that you decided to present this data visually on a map. I imagine some users would want to start with all cities selected and then exclude all cities not around a certain geographical point that they are interested in.

I like that you can get an overview and then more detail about a place. I’m not sure what the x1 and y1 stand for in the mouse over. Perhaps that data could be excluded or explained further? I also don’t think “prepared” is intuitive, and would suggest calling the filter “prepared meals” instead. I am curious as to why you decided to make prepared meals color-coded rather treating it as another filter? I think it works, I’m just not sure why that data is set apart in that way.

Finally, I agree that the table is a nice way to see a quick overview of different lists. I’m wondering if you could use a different color for those lists than green as green is already an indicator of prepared meals?

These are small suggestions for an overall impressive project. I appreciate the amount of data you are presenting in a coherent and concise space, and I think the tool would be useful for many people. Thanks for sharing!

Review 2: Madeline Coven

The visualization for “Number of farms, average farm size, % family-owned farms (per county)” is a wonderful use of map space for a great depth of data. The three levels are shown in ways that are prominent without being distracting. The blue squares for average farm size are easily separated visually from the circles for the percentage of family owned farms, and the total numbers of farms in counties, represented by the shapes of the counties. The three very different monochrome color schemes also allows distinction between the shapes even more easily, with oranges and yellows, greens, and one color blue with different sizes (different blues would have confused the data presentation).

This arrangement would lend itself well to a web-based interactive map that might allow visitors to explore the levels of data and their interactions by clicking on any data element easily and showing it in its own map or graph.

The ordinal presentation of information also allows the presentation to reveal all the desired information in a clear and clean way, putting a limit to the size of the squares for certain ranges of quantities. If the square sizes had been plotted on a continuous scale, it would have been too easy to make the squares too small to see, or so large that they obscured other information. Exact numbers might have produced differences that would have distracted from the message of the visualization.

Review 3: Rachel Mundstock

First off, congratulations, these are some wonderful and informative visualizations. I'm particularly struck by 'Task 3: Farmers markets and health and income factors' and 'Farmers market locations vs food deserts (by census tract).'

For all the visuals in Task 3 it is very easy to read the map and county lined because the bubbles are at an appropriate scale. It took me a second to grasp that the bubbles differed by size and color, since I immediately perceived the difference in size but not color. The color choice and gradation were used effectively. Since I'm not very familiar with North Carolina geography, the labeling of each county within Tableau was very helpful. My one suggestion would be to include some population information, or another line for number of farmer's markets per whatever number of residents.

The concept of food deserts has always been a bit hard for me to grasp since I've never been in such a situation. Marking the farmers markets and the food deserts was a fantastic idea, it's really clear that there are very few farmers markets near food deserts. I wish more of this type of visualization were available, it would have made it much easier to grasp the concept when I first learned about it. The visual communicates in a glance what a much longer explanation tries to achieve.

Review 4: Cheri Boisvert

This presentation looks very thorough. The maps are easy to interpret and include an interesting overlap of variables. The Number of farms, average farm size, % family-owned farms (per county) maps struck me as particularly innovative, covering three semi-related variables in one visualization without losing readability. The charts are also quite lovely. It took me a moment to interpret the County Data Correlations, but after a moment found it rather informative.

This is an interesting topic with local implications and something well worth exploring. I look forward to the presentation.

Review #4: Danny Nguyen

For Task 1, I liked how you used the Tableau application to map an interactive map that overlaid the Farmers Market data over a map of North Carolina. From the visual, I couldn’t tell if the counties were delineated with county lines, but that might’ve been useful to see to get a point of reference for the cities that were highlighted. The predominant white background might’ve been better represented as a light grey to lessen the harshness of the white on the eyes.

There seems to be a lot of filters and ways to manipulate the data to represent it on the map, which is think, makes it very customizable to the user’s needs; however, the multitude of choices is also a bit daunting upon initial inspection.

For Task 2, I liked how bar charts were used to show the Merchandise Available at Farmers Market and thought that it was a wise decision not to be tempted to complicate the visual. The Many Eyes treemap is a bit confusing to me and would take more time to decipher, since it is not intuitive to understand. There may be a better way to represent this data in a simpler manner.

For Task 3 and Task 4, I would say very similar things as Task 1. The visual in Task 4 would be better with the ocean shown as a blue color, but other than that, I think that the Task 1 and 3 should look visually similar to Task 4. The visual in Task 4 had a nice neutral grey background, showed the county lines, highlighted major cities, and was very visually appealing.

Reviewer Dr H

First kudos to your group for choosing and interesting problem, finding real wold data, and going the the challenging processing, cleansing, and merging of data to make our visualizations. It seemed like there were two main types of visualizations: descriptive (in terms of answer questions as Leondra and Carrie did), and visual analytics (attempting to understand the relationships between variables, particularly correlations, Pam and Sarah). Many good comments above. I'll add a few more.

For simple questions (like Leondra and Carrie's) the answers may be textual, or simple charts (like Carrie suggested). The challenge that I think you may be better equipped to address in your examples is understanding several of these together, in context. I.e. can you see who carries what food, takes credit cards, and is in driving distance from me (less than 20 minutes). Or similar. In this case the exploratory type tools like Leondra did in Tableau are effective I think. Note, it's not hard to tie in the driving distance part if you wanted to. Carrie, I liked your explorations, but not sure any of these strongly resonated with me. In particular the treemap where you're grouped on counties seems inherently less desirable than using a spatial map for the same purpose. I'd consider using treemap for grouping on some characteristic instead, and maybe include other characteristics or counties. (but even then, if I could do on a map that would be better).

Pam, I liked very much what you were attempting to do. My main comment is why use color of circles for "type" instead of color of counties? Usually for two distinct variables it's much easier for us to take them in that way. I.e. making a judgment about color of circles when circles vary in size (particularly when very small) is perceptually challenging. I.e. use something like from our class examples with size for farmers markets, and color of county (light pastel background) for child poverty rate. I liked that you did in tableau and included quick fliters, and that you used the dashboard! This makes exploration easy.

Sarah, liked the use of geocommons (and coverage of its limitations). I thought the visualizations were all good attempts, particularly when you tried to show 3 variables at once (using rectangles over county color plus the symbols for markets). This is tough. On these I liked your choice of intervals for the color coding-thought this made it stand out better (easily to get lost if shown as continuous). My favorite was your last one, where the chart visualization of correlations was really well done, and easily seen and comprehended. Nice grouping by main factors you are interested in studying. As discussed in class this would make excellent index into analytic side (visualizations like yours and Pam's with interactive filter like Pam had) for user to explore further. Personally, I'd like to see population included as well. I think if you could see several (3-4 variables at once, you could being to unravel the story).