Return to prices
Slide rule price analysis
As most regular users of the price data will have gathered, the price analysis is done using computer programs I have written myself. These notes give some background on how it is done, some warnings about limitations of the method and a lastly a few more details on the software used. The last section also covers a few other points relating to the web site.
How it's done
The raw data come from ebay, by doing a search on "slide rule" and the clicking on "search completed items". This usually give about 15 pages with 50 deals on each page covering the previous two weeks or so. I usually download the details three or four times a month and save the files in "txt" format - i.e. not as HTML format. This takes about 15 minutes each time - say an hour per month.
When the data have been downloaded I have a short program that reads all the files for a particular month, keeps only "unique" lines by doing a string compare, sorts them into date order and produces a new single file with all completed auctions for that month. It includes both those auctions where a sale was made and those where it was not.
Analysis for monthly sales
The analysis covers two areas: an overall one of sales ordered by value, number of bids, day of week and date; an analysis by manufacturer. It uses the file of all slide rule auctions for the month produced above. To extract the required information each line of text is "parsed". For example each line starts with 6 spaces and is followed by a 9 (usually) digit reference number. The item description is all the text between the number and the currency symbol and so on. After I have extracted all the information as text, appropriate values (e.g. price, number of bids, date, etc) are converted to numeric equivalents.
This provides the data needed for the overall analysis.
To analyse the sale by manufacturer I look for the "name" of the maker. To allow for different capitalisation I analyse the name after conversion to upper case. In some cases this is easy; for Pickett I look for the letters "PICKET" which will pick up the name with and without the "TT" at the end. In other cases it is more difficult: for Keuffel and Esser I look for nine different variants based on the name and combinations of "K & E", "K&E" etc.
After analysis the program writes the results directly as HTML files, with all the relevant header and table formatting text.
Analysis of time trend
To analyse the trend over the time I concatenate the file of the prices for the latest month at the end of a file made up of all the previous files of data for a single month. (Using the DOS command of the form "Copy OldCompletefile+Newfile NewCompleteFile). This takes each day's completed auctions and ranks them in price order. From this data I get the average and various quantiles. For example if there were 24 sales I take the 6 highest prices as the upper quartile; in reality the 6th highest price is used for the upper quartile for 24 to 27 sales, I don't do anything sophisticated like weighting between the 6th and 7th values when the number of sales in not divisible by 4.
These values are output as a "csv" file which I then import to Excel. The plotting and moving average values are done in the spread sheet. One of the most fiddly steps is getting the graph into an appropriate format for the web site. By default the software I use converts the graph to JPG format (which is good for photos but not lines) so I have to export it to Power Point, save it as a GIF file (which is good for lines and not photos) and import it the web software.
Other analyses, such as variation in price through the week and sales of popular models, were done using separate programs.
Some limitations and warnings
Not all sales are completed in US dollars, however the completed sales file uses only the dollar currency sign ($) whatever currency was used for the original transaction. For example, if a sale was completed for 30 pounds sterling it becomes $30.00 in the completed sales file.
Even though the analysis is based on uniquely identifiable fragments of the names of makers it is not possible to allow for the variety of spellings used by sellers. There is a separate file for all makers apart from the 18 I identify and some of the "identified" makers may find their way there.
There is a number of other possible limitations which users should be aware of:
- Some sellers bundle more than one rule as lot. This would be treated as one rule by
the first identified maker.
- Some manuals are sold separately. These would be treated as slide rule.
- Words which appear frequently in the item description but which are also a maker's name, e.g. Unique, are treated simply as adjectives.
- The search is based on "slide rule". Some rules are sold as "slide-rule", "sliderule" and other variants. These probably constitute about 5% of sales but are not analysed.
The programs are written in Visual Basic 6.
The Web Page is designed using Front Page 98.
The images are prepared using Micrografx Picture Publisher and scanned using an IBM Idea Scan. All this images are scanned at 150 dpi. Details of the rules are kept at this resolution. Images of complete rules are first prepared at this resolution. Most of the rule are too long to use the scanner "longways" so the rules are scanned twice "sideways" and the images are joined at the edge of the cursor where shadows and blurring of the image hide the join. The image of the complete rule is reduced to 50 dpi. I feel it is important to maintain scale consistency as that way the relative size of different rules can be appreciated. The images are saved as jpg files with a relatively low compression and using an option available with Micrografx to preserve detail.
The text of manuals was scanned at 300 dpi and converted to text by Textbridge Pro 98.
Return to prices