The city of Seattle makes available its database of pet licenses issued from 2000 to the beginning of 2020 as part of the city's ongoing Open Data Initiative. This post will explore extracting the data from Seattle's Open Data portal using requests, then transform the extracted JSON data into a workable dataset with pandas to analyze and investigate the pet license database.

About Seattle Pet Licensing¶

The city of Seattle requires pets over eight weeks old be licensed. There are several benefits to licensing one's pet, including a return ride home if your pet is lost, and easier contact from a veterinarian if your pet is unfortunately injured. If the licensing is performed at the Seattle Animal Shelter on the third Saturday of any given month, a free rabies vaccine is included, as well as other vaccines and a microchip for a small additional fee.

Getting Started¶

Import the %matplotlib inline magic function for displaying the output visualizations inline in the notebook and import the libraries that will be used in this section of the analysis.

Here, we set the formatting for the plots created with seaborn that appear later in the analysis.

Getting the Data¶

We extract the data programmatically from the Seattle Open Data Portal. The portal utilizes Socrata to host its data. Therefore we will be working with the Socrata API. The Seattle pet licenses database API endpoint can be found on the dataset's information page by clicking the API button in the top-right corner.

Socrata's paging documentation states there is a return limit of 1,000 for each call. Therefore, we need to page the results to capture the entire dataset. The default format returned from the API is JSON, though other formats such as CSV and XML are also available. We select the default JSON format and collect the results into a list that we can later use to build a pandas DataFrame. The following block extracts the data from the API into a list until there are no more results to return.

The data is now stored in a list that we can iterate through and normalize each JSON result using pandas's handy json_normalize function. Once all the JSON data is collected into a DataFrame, we can inspect the number of rows and print the first few rows of the data.

The row count is just over 49k, which is the same number of rows given on the dataset's portal page. We see the data has seven columns, and are the same as listed on the data's webpage. We now have a nice, tidy dataset that was created programmatically with a combination of libraries! Isn't that much more satisfying than just downloading the data as a .csv from the website? (Of course, it is).

It is often a good idea to save the extracted and transformed data onto the disk to avoid having to repeatedly get the data from the API if it is to be used later. We quickly save the DataFrame as a .csv using the to_csv method.

Exploratory Data Analysis of the Seattle Pet Licenses Dataset¶

Now that we have a tidy dataset to work with, we can move onto exploring the data! As a start, let's see how many pet licenses have been issued by species of pet.

We see a vast disparity, more than 2 to 1, between the count of licenses issued to dogs and cats. Some Seattle residents have registered goats and pigs as pets as well, though, as one may expect, the volume of registrations for goats and pigs is far lower than more common household pets.

According to a survey conducted by the American Pet Products Association conducted between 2017 and 2018, there were an estimated 94.2 million cats, averaging two cats per household and an estimated 89.7 million dogs averaging just under 1.5 dogs per household. Assuming these numbers are proportional to the pet-owning population in Seattle, why are dogs licensed almost twice as much as cats? Some possible factors include:

• One of the main benefits of getting a license for one's pet is the pet can be returned at no cost should the pet, unfortunately, go missing. As dogs are often outdoors much more frequently than cats for walks and such, perhaps the thought of one's dog becoming lost is more prevalent in dog owners' minds.
• According to another survey conducted in 2012 by the American Veterinary Medical Association, 66.7% of dog owners surveyed thought their dog(s) are a part of their family, compared to 56.1% of cat owners surveyed. The perception of family inclusiveness could be another factor for the difference in licensing rates as it can be reasonably assumed that owners who consider their pets part of their family would be more likely to purchase a license.
  • The same AVMA survey also found that the average amount spent on veterinary care for each dog per year was $227 compared to just $90 for cats. Although this could be further evidence of the general attitude towards dogs and cats as family members, it is typically more challenging to determine if a cat is ill as they are not prone to showing they are not well, therefore this could also play a role in the difference in amount spent on care.

As we are more interested in dog and cat pet licenses for this analysis and the low overall volume of licenses issued to other types of pets, we remove the goat and pig licenses from the data.

Finding Most Common Dog and Cat Names¶

The pet's name for each license is also available in the licenses dataset. Using this information, can we find any similarity in the names of cats and dogs given by their owners? To answer this question, let's first plot the top ten cat and dog names by total count.

The most common name for both dogs and cats is Lucy, with the names Max, Luna, Bella, Charlie, and Max, also reaching the top 10 names for both animals. Given that 60% of the top ten cat and dog names are the same between animals, we can reasonably assume there is a decent amount of commonality in names amongst dogs and cats.

Now let's examine the 50 most common cat and dog names by total count and see the intersection of names using some of Python's available set operations. We print the names that are shared by cats and dogs in the top 50 names, as well as the total number.

Interestingly, 29 of the 50 top cat and dog names are shared, almost precisely the same ratio (60%) found in the list of top ten names.

Determining the Most Common Pet Breeds¶

According to the American Kennel Club, Labrador Retrievers, German Shepherds, and Golden Retrievers are the most popular dog breeds of 2016 and have been since 2013. The top three most popular cat breeds, according to a recent article by The Spruce are the Siamese, Persian, and Maine Coon. Using the primary breed information available in the pet license database, we can see if dog and cat owners in Seattle confirm the results as reported by the articles.

Here, we plot the top 20 most frequently appearing primary breeds of dogs and cats.

Based on the available data, we see the top three dog breeds are Labrador and Golden Retrievers and Short Coat Chihuahuas, with German Shepherds and Terriers making up the fourth and fifth most popular breeds. Technically speaking, domestic cats are not considered a pure breed, as they are the mix of several different breeds. Excluding the domestic breeds, we see the next most popular cat breeds are the American Shorthair, Siamese, and the LaPerm. The American Shorthair is the twelfth most popular breed in The Spruce's list while the LaPerm doesn't make an appearance.

Dog breeds also have a more even distribution while cat breeds are heavily skewed towards the domestic breeds, particularly the short hair variety. There are many potential reasons for this, of which the primary one is cat breeding is relatively new in comparison to breeding dogs for specific traits. Thus there are far fewer cat breeds recognized by the relevant governing bodies than dogs. It is also much more challenging to identify the breed of a cat, the only surefire way being genetic testing.

Analysis of Pet Licensing by Geography¶

The zip code of the pet's home is also given in the dataset. As the data is user-generated, we need to perform a few cleaning steps to get the zip codes all in a similar format before we can analyze the data further.

Zip codes are often given a four-digit suffix beginning with a hyphen for more specificity regarding the particular area. This suffix is not required for our purposes. Therefore we convert the zip code into a string and split the data by the hype and keep only the first piece.

Now that we have a cleaner set of zip codes, we can start by plotting which zips have the most licensed pets. As there are quite a lot of zip codes, we focus on the top 15 by volume. A count column filled with $1$s is also added for convenience when plotting.

Only three zip codes, 98115, 98103, and 98117 have more then 3,000 licensed pets. Anecdotally, the fact that the 98115 zip code has the most licensed pets doesn't surprise as I lived in the 98115 zip code for several years and could always see numerous dogs being walked throughout the day.

As one would reasonably expect, given the knowledge of an almost 2:1 ratio in the number of cats licensed compared to dogs, the top 15 zip codes by total licensed pets also shown a significant difference in licensed dogs and cats.

Finding Popular Dog and Cat Breeds by Zip Code¶

Earlier in the analysis, we found the most frequently appearing cat and dog breeds in the pet licenses database. Does the frequency of breeds appearing differ when viewed at a zip code level? That is, are there areas of Seattle that favor a particular dog or cat breed over others? One could reasonably hypothesize that smaller dogs such as Chihuahuas, Miniature Poodles, and Pugs would be more common in the denser areas of the city due to space and that some apartment buildings charge extra for larger breeds of dogs.

The following creates two new data frames that contain the top five cat and dog breeds for the top five zip codes. A rank column is added to find the most frequently occurring breeds by each zip code using pandas's groupby and rank() methods that are then used to filter the breeds by the top five.

Plot the top five most frequently occurring dog and cat breeds within the zip codes with the most licensed pets.

As one might reasonably suspect, the most common breeds overall are also the most frequently appearing within each zip code, although there are several ties for the fifth most popular dog breed. Is there any more variety when the domestic cat breeds and retriever dog breeds are removed from the results before ranking?

After the removal of the domestic cat breeds, we see there is a relatively even distribution of several of the most popular breeds, including the Siamese, American Shorthair, LaPerm, and the Maine Coon. The frequency of breeds is also somewhat consistent across zip codes, with Siamese and American Shorthairs being the most popular, followed by LaPerms and Maine Coons. Could these results be indicative of a general breed preference amongst Seattle cat owners? Given the comparatively low frequency of these breeds appearing in comparison to the domestic breeds, there likely isn't much of a relationship; however, it could be interesting to compare these results with a survey or some other data set that gave cat owner breed preferences. Similar to the Siamese cat, the Short Hair Chihuahua is the most popular dog breed after the exclusion of Labrador and golden retrievers. What is perhaps most interesting is Miniature Poodles only appear in the 98115 and 98117 zip codes, as well as Beagles only showing up in the 98103 zip code.

Time Series Analysis of Issued Pet Licenses¶

The Seattle pet licenses dataset contains the date the license was registered with the city, which gives us the ability to investigate how the volume of licenses being issued has changed over time. To begin analyzing the time series, we first convert the datasets issue_date column into a datetime format using pandas's to_datetime function.

Plot the number of licenses issued by day for all the days included in the dataset.

There appears to be very little to no pet licenses being issued from the beginning of 2000 up until about mid-2013 when some noticeable volume began to emerge. It isn't until near the end of 2014 that the number of issued pet licenses balloons to over a hundred or more each day.

What could cause such an extreme shift in the volume that has continued for just over three years after almost no activity for close to 10 years? Some cursory searching on Google News did not return anything definitive that could be a possible factor of the sudden change, nor did any further investigation on the Seattle Animal Shelter website.

To get a better visualization of how the data appears to trend, split the data by pre-2015 and post-2015, and plot.

The data plotted separately further displays the stark difference in pet licensing volume that began in November of 2014 and has continued as far as the data is available. Post-2015, it seems there are a few peaks at the end of each year around November and December. Other than that, there doesn't appear to be any significant trend.

We can explore the trend and other components of the time series by performing time series decomposition. Decomposing a time-series into unique components that each represent a particular facet of the time series data helps identify any cyclical or seasonal trends and is an essential piece of any time series analysis.

The statsmodels library provides a function [seasonal_decompose] that performs decomposition on a time series. The frequency parameter freq is set to 12 to indicate monthly frequency, though other values such as 7 (weekly frequency) or 3 (quarterly frequency), could also be used.

To get a more visually appealing graph of the decomposed time-series, we access the attributes that represent the fitted components and plot using pandas's convenient visualization methods.

There does appear to be a trend judging by the component resulting from the decomposition. Ups and downs show throughout the year and then rising significantly near the end of the year. The seasonal component is minimal, thus there likely isn't much, if any, seasonality. To get a better sense of the trend, we can overlay the component with the original time series.

Before moving into the next section of the analysis, which focuses on the issuing of pet licenses by particular zip codes in Seattle, let's see if there is any significant difference in the volume of licenses issued depending on the day of the week.

Do Higher Income Seattle Areas License Their Pets More Frequently Than Other Areas?¶

Earlier in the analysis, we noted that only three zip codes have more than 3,000 licensed pets. Is there any likely cause or causes behind the stratification of licensed pets within the city of Seattle? One possible reason may be related to the demographics of the neighbors the zip codes encompass. For example, do more affluent areas of Seattle tend to have a higher amount of licensed pets?

Using data provided by the IRS, we can analyze the number of tax returns received by different ranges of income and determine if there is any correlation between the number of pets licensed through the city. The data is available from the IRS Statistics of Income page. Clicking on an individual state on the map or from the list of states downloads an Excel file with a range of tax return statistics by income brackets. The most recent data available is from 2015, which is current enough for our purposes. In its raw form, the data is somewhat messy and hard to analyze. Therefore a cleaner copy in CSV format with only the data we need was created to facilitate the merging and analysis of the data. The CSV copy of the data can be downloaded from an AWS S3 bucket. We load the data now using pandas and print the first few rows to check out the data.

Let's quickly clean up the Adjusted Gross Income column by replacing 'under' with a hyphen. We also create a list of the column names that will be used later for changing the order of appearance of the income brackets in the visualization. The raw number of returns is transformed into a percentage of the total count of returns made within each respective zip code to see how the zip codes vary by the proportion of income brackets.

The majority of tax returns are from those making approximately $75,000 or less. This majority is even higher in some zip codes, such as 98118 and 98105, reaching nearly 60% of IRS filers. The 98199 zip code appears to be the most affluent based on data available according to the proportion of tax returns made by registrants with over $100,000 in income.

Earlier, we saw the 98115, 98117, and 98103 zip codes have the most licensed pets. Based on the tax return data, these zip codes appear to be more affluent than others, but not by much. The 98199 zip code is in the middle regarding total licensed pets. With this knowledge, we can cast some doubt that there is a high correlation between the number of licensed pets and the affluence of the respective neighborhood.

Using pandas's handy corr() method and seaborn's heatmap visualization function, we can construct a correlation matrix to see how the number of licensed pets correlates to the number of tax returns by various income brackets. To accomplish this, we must merge the two datasets using the zip code as the key. Pandas's merge() function is quite useful in combining data sets based on similar keys with a SQL-like interface.

The correlations to the number of returns by income bracket and total pet licenses can be found in the first column and first row of the correlation matrix. We see the correlations are not strong in the absolute sense (close to 1) but could be considered high or significant as the data is social and economic. Interestingly, the income brackets with the highest correlations to pets licensed are in the $25,000 to $75,000 range, while the larger income brackets, $100,000 to more than $200,000, are slightly lower.

Based on these results, it seems there could be something of a relationship between the number of licensed pets and the wealth of the particular area, but if there is it is most likely quite weak. Another method of finding any possible relationships in the data is to perform linear regression, a standard model that fits a linear equation to a dependent variable and one or more independent variables. Although linear regression in its purer forms is comparatively straightforward to more exotic approaches, the technique is still quite powerful and is very much in use today due to its relative ease of use and interpretability, as well as the amount of information that can be gleaned from inspecting the model.

Several popular Python libraries can perform linear regression, of which include scikit-learn's LinearRegression, numpy's lstsq, and Statsmodels' OLS, all of which are different and offer various advantages. For our purposes, we are more interested in the model's diagnostics and tests to see if we can find any additional useful information for identifying any relationship between the frequency of pet licensing and the affluence of individual neighborhoods. The Statsmodels implementation provides a very lovely R-like summary of the linear regression model fit, which fits nicely for our needs.

To fit the linear model using the Statsmodels library, we first create the x (independent) and y (dependent) variables. According to the OLS() documentation, a constant that represents the intercept must be added using add_constant before fitting the model.

The Statsmodels summary() method prints a nice table of the linear regression fit results. We see the coefficient of determination, better known as $R^2$ is slightly under $0.50$ and a p-value well below $0.05$. Looking at the individual model coefficients, several income range coefficients have a p-value below 0.05.

Based on these results, it seems there could be something of a relationship between the number of licensed pets and the wealth of the particular area, but if there is, it is most likely quite weak. Another method of finding any possible relationships in the data is to perform linear regression, a standard model that fits a linear equation to a dependent variable and one or more independent variables. Although linear regression in its purer forms is comparatively straightforward to more exotic approaches, the technique is still quite powerful. It is very much in use today due to its relative ease of use and interpretability, as well as the amount of information that can be gleaned from inspecting the model.

Conclusion¶

We explored several facets of the Seattle Pet Licenses dataset in this analysis and found several interesting pieces of information that could be useful in a variety of different ways. One possible use of such information is planning a marketing or outreach campaign to target cat owners who haven't considered licensing their cat in zip codes with a higher disparity between the number of cats and dogs. What follows is a summary of some of the new information we learned from performing this analysis:

• The most popular dog and cat name in Seattle, according to data available, is Lucy. 29 of the top 50, or nearly 60%, of dog and cat names, are shared.
• There are only three zip codes with more than 3,000 licensed pets. We saw there isn't much of a relationship between the number of licensed pets and the distribution of wealth in a particular zip code. Therefore other factors contribute to the number of licensed pets.
• The number of pets licensed appears to trend upwards near the end of the year, specifically the fourth quarter. This trend has occurred for the past two years, so it will be interesting to see if it continues through 2020 and on (should data become available).
• Dogs are licensed almost twice as much as cats, although there are an estimated five million more pet cats in the United States. It is difficult to say conclusively why this would be without more data; however, one could reasonably assume one factor is a misconception that a dog is more likely to go missing than an indoor cat.
• The correlation between the number of licensed pets and the distribution of wealth in a particular area according to the volume of tax returns for various income brackets could be considered strong given the social context of the data; however, this does not imply any causation.