Using Spotfire Visualization Tool for Laptop Purchase Research

Introduction

It is commonplace now for consumers to do research on internet before they buy some expensive merchandise. In this case, I was buying a laptop computer for a friend. I chose to use www.cnet.com's hardware site for preliminary research. Downloaded and cleaned the data, and used Spotfire as a navigation tool to study the data.

Description of the Data

A list of 246 kinds of Celeron, AMD K6-2, AMD K6-III based computers have specification information on cnet.com. Important features of a laptop computer include model, maker, cpu type, cpu clock speed, amount of memory, size of hard drive, LCD screen size and price. (To download this data set, click here.)

The importance of these features, however, are usually not equal for a laptop buyer. For example, CPU clock speed is increasing every quarter, therefore a 400 MHz machine is not much different from a 433 MHz one. Whereas LCD screen size usually makes a big difference in price (e.g. $400 difference between a 12.1 inch and a 14.1 inch).

The Visualization

The online version of the data display is in pages after pages of tables with links to online resellers or manufacturer sites. Usually, these web sites categorize the data in a few different ways. You can browse by price range, or by cpu type, or by manufacturer. You can also make complex queries to get a small table of results matching your query parameters. (See cnet site.)

Dynamic-query based graphical visualization systems, such as Spotfire, does make a difference in presenting the data.

Spotfire has a lot of advantages over long html tables:

it gives an overview of hundreds of data items in one screen; This screen snapshot shows all 246 computer models. X-axis shows price, Y-axis shows CPU clock speed. Color-coding is used for CPU type: red for celeron, blue for AMD K6-2, yellow for AMD K6-3. Size coding is for LCD screen size. 4 of the important features are all displayed.
dynamic query enables quick filtering of query results; This screen snapshot shows the results of limiting all machines to Compaq with either Celeron or AMD K6-III chip.
comparisons among very different machines/configurations are easy and fast; This screen snapshot shows the comparison of Compaq, Dell, and Gateway laptops. In the 3-d visualization, a new axis for amount of RAM is added.

easy to find outliers and anomalies; Most of the machines in this set are $2,500 or less ("budget notebooks" as opposed to PII based "power systems"). However, there are 4 models with relatively slow clock speed, small screen size, and price way out of order. I really don't think they are going to sell well:

      Model	                        CPU	ClockSpeed	RAM	HardDrive	Display	Price
      Fujitsu Biblo MF33		Celeron	333      	64      6.4      	12.1	$2895
      Panasonic Let's Note S51 EX	Celeron	300     	64	6.4     	11.3	$3395
      Panasonic Let's Note S51 V	Celeron	300     	64	4.3     	11.3	$3395
      Sharp Mebius S3           	Celeron	333     	64	6.4     	11.3	$3195

trend and anomalies: click to see full sized image

Weakness of Spotfire

For consumer purchase research purpose, Spotfire still has some problems. Improvements are desired in the following ways:

A Java version that can run on the web would be cool;
Although Spotfire offers a number of ways to customize the graphic output, it is still designed as a general tool to work on "any" general data. I find the graphics is still rather limited and unattractive. Only a few 2-d, 3-d shapes are supported. In some cases, the graphics support in Spotfire is not very user-friendly. The graphs make sense to trained statisticians, may not make much sense for general public who would use such a research tool most for purchase purpose. More sophisticated and customizable graphics support may be introduced;
Screen usage is still not efficient. A large portion of the screen real estate is blank. On the other hand, query device (sliders, etc) window is rather small by default. Detailed window often can not be shown fully. Microsoft Windows is also to blame: just look at how many pixels those sliders have used:) Scatter plots can be made with even higher resolution and smaller. Zooming may be used to reduce the size of overview thus creating more room for query devices and detail/query result windows.

Conclusion

To use graphical dynamic query tools such as Spotfire in consumer purchase research would bring a lot of convenience and insights to the research. If it can be made runnable on web, and more customized to the particular data domain, it will be a very powerful and attractive tool. I think there is need for such a commercial product for the ever-growing e-commerce sites.

Jin Tong
Last modified: Mon Oct 04 23:18:01 Eastern Daylight Time 1999

Web Accessibility