Sorry, an error occurred while loading the content.

## Clustering US lawmakers by the similarities of their voting records

Expand Messages
• Using data from govtrack.us, I wrote this article: Visualized grouping of US Congress members by similarities in their voting records Visualized grouping of US
Message 1 of 5 , Jun 19, 2014
Using data from govtrack.us, I wrote this article:

## Visualized grouping of US Congress members by similarities in their voting records

 Visualized grouping of US Congress members by similariti...The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014. View on waliberals.org Preview by Yahoo

The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014.
What’s most notable is how different the Dems are from the Repugs.

The data on which this analysis is based come from https://www.govtrack.us/.   The analysis and clustering are by me. A handful of lawmakers with few votes are omitted.

Please realize that the two dimensional images below are projections from a high-dimensional space of data, and so there is loss of information. The images represent approximate summaries.  Also realize that the x and y directions in these images have no particular meaning: the images show legislators placed in such a way as to optimize their relative distances, and this is accomplished by a stochastic relaxation algorithm.

## Section 1: US Senators clustered by the similarity of their voting records in 2013 and 2014

Click on the images to see detail.

Notice that the Democrats are together on the upper-right, and the Republicans are spread out on the bottom-left. It seems that Republicans vote less as a block than the Democrats in the Senate.

## Section 2: US Representatives clustered by the similarity of their voting records in 2013 and 2014

Click on the image to see detail.

...

Don Smith

• Repugs? Could you please be a little more tendentious? Also: could you say something about your methodology? Is this the result of K-Means with randomly
Message 2 of 5 , Jun 21, 2014
Repugs? Could you please be a little more tendentious?

Also: could you say something about your methodology? Is this the result of K-Means with randomly initialized centroids, or is this just the way the visualization falls out from positioning each member of Congress in the vote space? If the latter, are the distances between the two parties on the same scale as the distances within a party?

Chuck

On Thu, Jun 19, 2014 at 7:29 PM, 'Donald A. Smith' thinkerfeeler@... [govtrack] wrote:

Using data from govtrack.us, I wrote this article:

## Visualized grouping of US Congress members by similarities in their voting records

 Visualized grouping of US Congress members by similariti... The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014. View on waliberals.org Preview by Yahoo

The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014.
What’s most notable is how different the Dems are from the Repugs.

The data on which this analysis is based come from https://www.govtrack.us/.   The analysis and clustering are by me. A handful of lawmakers with few votes are omitted.

Please realize that the two dimensional images below are projections from a high-dimensional space of data, and so there is loss of information. The images represent approximate summaries.  Also realize that the x and y directions in these images have no particular meaning: the images show legislators placed in such a way as to optimize their relative distances, and this is accomplished by a stochastic relaxation algorithm.

## Section 1: US Senators clustered by the similarity of their voting records in 2013 and 2014

Click on the images to see detail.

Notice that the Democrats are together on the upper-right, and the Republicans are spread out on the bottom-left. It seems that Republicans vote less as a block than the Democrats in the Senate.

## Section 2: US Representatives clustered by the similarity of their voting records in 2013 and 2014

Click on the image to see detail.

...

Don Smith

• Also, I haven t done much with visualizations, so if you could talk about the tools and algorithms you used to flatten the voting distances, that would be very
Message 3 of 5 , Jun 21, 2014
Also, I haven't done much with visualizations, so if you could talk about the tools and algorithms you used to flatten the voting distances, that would be very helpful.

Thanks,
Chuck

On Sat, Jun 21, 2014 at 8:26 AM, Chuck Bearden wrote:
Repugs? Could you please be a little more tendentious?

Also: could you say something about your methodology? Is this the result of K-Means with randomly initialized centroids, or is this just the way the visualization falls out from positioning each member of Congress in the vote space? If the latter, are the distances between the two parties on the same scale as the distances within a party?

Chuck

On Thu, Jun 19, 2014 at 7:29 PM, 'Donald A. Smith' thinkerfeeler@... [govtrack] wrote:

Using data from govtrack.us, I wrote this article:

## Visualized grouping of US Congress members by similarities in their voting records

 Visualized grouping of US Congress members by similariti...The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014. View on waliberals.org Preview by Yahoo

The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014.
What’s most notable is how different the Dems are from the Repugs.

The data on which this analysis is based come from https://www.govtrack.us/.   The analysis and clustering are by me. A handful of lawmakers with few votes are omitted.

Please realize that the two dimensional images below are projections from a high-dimensional space of data, and so there is loss of information. The images represent approximate summaries.  Also realize that the x and y directions in these images have no particular meaning: the images show legislators placed in such a way as to optimize their relative distances, and this is accomplished by a stochastic relaxation algorithm.

## Section 1: US Senators clustered by the similarity of their voting records in 2013 and 2014

Click on the images to see detail.

Notice that the Democrats are together on the upper-right, and the Republicans are spread out on the bottom-left. It seems that Republicans vote less as a block than the Democrats in the Senate.

## Section 2: US Representatives clustered by the similarity of their voting records in 2013 and 2014

Click on the image to see detail.

...

Don Smith

• Hello, Chuck, your emails somehow ended up in my spam folder. Hence my delay in responding. The distance between lawmakers is equal to how many votes (rolls)
Message 4 of 5 , Jun 23, 2014
Hello,

Chuck, your emails somehow ended up in my spam folder. Hence my delay in responding.

The distance between lawmakers is equal to how many votes (rolls) they voted differently on. If two lawmakers had the exact same votes, they'd be at distance zero.  So, the Dems and Repubs are on the same scale.

Perhaps I should exclude certain kinds of votes (e.g., procedural votes); or classify the votes by category or issue; that would expose other info about lawmakers.

I did the clustering by a stochastic relaxation algorithm that randomly tweaks the location of lawmakers, with decreasing sizes of tweaks, choosing configurations that minimize the cost. It's like natural selection.

I loaded the weighted graph of distances into Gephi, the graph visualization and analysis toolkit, and the 2d graphs that Gephi produced looked much like the ones I generated with my stochastic relaxation algorithm.

I wrote a follow-up article:

## Moderate Republican Senators Collins and Murkowski voted more similary to Dems than to Repugs

If you know of related visualizations, please point me to them.

Thanks, Don

 Gephi makes graphs handlyApplications Exploratory Data Analysis: intuition-oriented analysis by networks manipulations in real time. View on gephi.gi... Preview by Yahoo

On Saturday, June 21, 2014 6:58 AM, "Chuck Bearden cfbearden@... [govtrack]" <govtrack@yahoogroups.com> wrote:

Also, I haven't done much with visualizations, so if you could talk about the tools and algorithms you used to flatten the voting distances, that would be very helpful.

Thanks,
Chuck

On Sat, Jun 21, 2014 at 8:26 AM, Chuck Bearden wrote:
Repugs? Could you please be a little more tendentious?

Also: could you say something about your methodology? Is this the result of K-Means with randomly initialized centroids, or is this just the way the visualization falls out from positioning each member of Congress in the vote space? If the latter, are the distances between the two parties on the same scale as the distances within a party?

Chuck

On Thu, Jun 19, 2014 at 7:29 PM, 'Donald A. Smith' thinkerfeeler@... [govtrack] wrote:

Using data from govtrack.us, I wrote this article:

## Visualized grouping of US Congress members by similarities in their voting records

 Visualized grouping of US Congress members by similariti...The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014. View on waliberals.org Preview by Yahoo

The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014.
What’s most notable is how different the Dems are from the Repugs.

The data on which this analysis is based come from https://www.govtrack.us/.   The analysis and clustering are by me. A handful of lawmakers with few votes are omitted.

Please realize that the two dimensional images below are projections from a high-dimensional space of data, and so there is loss of information. The images represent approximate summaries.  Also realize that the x and y directions in these images have no particular meaning: the images show legislators placed in such a way as to optimize their relative distances, and this is accomplished by a stochastic relaxation algorithm.

## Section 1: US Senators clustered by the similarity of their voting records in 2013 and 2014

Click on the images to see detail.

Notice that the Democrats are together on the upper-right, and the Republicans are spread out on the bottom-left. It seems that Republicans vote less as a block than the Democrats in the Senate.

## Section 2: US Representatives clustered by the similarity of their voting records in 2013 and 2014

Click on the image to see detail.

...

Don Smith

• In fact, it s not really a clustering algorithm, in the sense of K-means. It s more like multidimensional scaling or graph layout. I edited the article
Message 5 of 5 , Jun 23, 2014
In fact, it's not really a clustering algorithm, in the sense of K-means. It's more like multidimensional scaling or graph layout. I edited the article

to reflect this fact.  And I made it less tendentious.

Don

On Monday, June 23, 2014 9:03 PM, Donald A. Smith <thinkerfeeler@...> wrote:

Hello,

Chuck, your emails somehow ended up in my spam folder. Hence my delay in responding.

The distance between lawmakers is equal to how many votes (rolls) they voted differently on. If two lawmakers had the exact same votes, they'd be at distance zero.  So, the Dems and Repubs are on the same scale.

Perhaps I should exclude certain kinds of votes (e.g., procedural votes); or classify the votes by category or issue; that would expose other info about lawmakers.

I did the clustering by a stochastic relaxation algorithm that randomly tweaks the location of lawmakers, with decreasing sizes of tweaks, choosing configurations that minimize the cost. It's like natural selection.

I loaded the weighted graph of distances into Gephi, the graph visualization and analysis toolkit, and the 2d graphs that Gephi produced looked much like the ones I generated with my stochastic relaxation algorithm.

I wrote a follow-up article:

## Moderate Republican Senators Collins and Murkowski voted more similary to Dems than to Repugs

If you know of related visualizations, please point me to them.

Thanks, Don

 Gephi makes graphs handlyApplications Exploratory Data Analysis: intuition-oriented analysis by networks manipulations in real time. View on gephi.gi... Preview by Yahoo

On Saturday, June 21, 2014 6:58 AM, "Chuck Bearden cfbearden@... [govtrack]" <govtrack@yahoogroups.com> wrote:

Also, I haven't done much with visualizations, so if you could talk about the tools and algorithms you used to flatten the voting distances, that would be very helpful.

Thanks,
Chuck

On Sat, Jun 21, 2014 at 8:26 AM, Chuck Bearden wrote:
Repugs? Could you please be a little more tendentious?

Also: could you say something about your methodology? Is this the result of K-Means with randomly initialized centroids, or is this just the way the visualization falls out from positioning each member of Congress in the vote space? If the latter, are the distances between the two parties on the same scale as the distances within a party?

Chuck

On Thu, Jun 19, 2014 at 7:29 PM, 'Donald A. Smith' thinkerfeeler@... [govtrack] wrote:

Using data from govtrack.us, I wrote this article:

## Visualized grouping of US Congress members by similarities in their voting records

 Visualized grouping of US Congress members by similariti...The following images show US Senators and Representatives clustered according to the similarity of their voting records in 2013 and 2014. View on waliberals.org Preview by Yahoo

The following images show US Senators and Representatives clusteredaccording to the similarity of their voting records in 2013 and 2014.
What’s most notable is how different the Dems are from the Repugs.

The data on which this analysis is based come from https://www.govtrack.us/.   The analysis and clustering are by me. A handful of lawmakers with few votes are omitted.

Please realize that the two dimensional images below are projections from a high-dimensional space of data, and so there is loss of information. The images represent approximate summaries.  Also realize that the x and y directions in these images have no particular meaning: the images show legislators placed in such a way as to optimize their relative distances, and this is accomplished by a stochastic relaxation algorithm.

## Section 1: US Senators clustered by the similarity of their voting records in 2013 and 2014

Click on the images to see detail.

Notice that the Democrats are together on the upper-right, and the Republicans are spread out on the bottom-left. It seems that Republicans vote less as a block than the Democrats in the Senate.

## Section 2: US Representatives clustered by the similarity of their voting records in 2013 and 2014

Click on the image to see detail.

...

Don Smith

Your message has been successfully submitted and would be delivered to recipients shortly.