Statistics (LozStats) Manual Version 2.41 DISTRIBUTION: Statistics version 2.41 is public domain emailware. Anyone can use it and you are free to distribute it - as long as you're not making pots of money out of it! If you use it you should send me a woman ;) or an email. INTRODUCTION: I can only remember seeing one other statistics package on the Amiga and that was in the early days (around 1987) and it cost several hundred pounds. Although a number of spreadsheets are available, none compare with Microsoft Excel or Lotus 1-2-3 for (ugh!) Windows and spreadsheets are not that easy to use for statistical analysis. This version allows 50 variables and 500 cases. Unfortunately it's still a bit rough around the edges. Considerable development time was wasted because of a bug in HiSoft Basic 2.00 that HiSoft were unwilling to fix. As a result I had to change to Blitz Basic 2.10 which is faster, but has required a massive conversion effort. I do not intend to explain the statistical theory of the tests - if you do not understand what the program is for - this package is not for you. A good easy to understand statistics book to accompany the program would be: "Business Statistics, A Decision Making Approach" (1985) D.F. Groebner & P.W. Shannon, published by Merrill. I'm sorry but I accept no responsibility for any difficulties (crashes, inaccurate results etc.) you encounter, as a result of using this software. If you have any problems drop me a line and I'll see what I can do. Contact .............. Dr. Lawrence Harrell BNR Consulting Ltd., bnr.cons@virgin.net INSTALLATION AND REQUIREMENTS: The program can be installed on hard disk or copied to your own floppy disks. Please note that the program requires the Topaz 11 font to be present in the Workbench "Fonts" directory. Topaz 11 is provided on all Amiga systems. Workbench 2.00 or above is required. I have no idea of the memory requirements, as I have 34mb, but I think that it will run on a 2mb machine. BASIC OPERATION Upon loading the program you are presented with the data entry screen. If you have a compatible statistics file that you wish to work on, select "load" from the pull down menu and load the file. If you are creating a new file, select "add variable" from the pull down menus. Continue until all variables have been entered. Next select "add cases" and type in the number of cases in your data set. If you are not sure how many cases there are - type in an estimate (it is easy to add or delete cases later). Your data entry template is now complete. Until the data has been entered all cells are filled with the missing value - cells containing missing values are ignored in the analysis. You can select the number that represents the missing value from the pull down menus. The default value is 99. To enter/edit data move the cursor to the cell you wish to change using the arrow keys. Now type the number and move to another cell. The number is automatically entered when an arrow key is pressed (you do not need to press enter). NOTE that the number you have typed will not appear on the screen until an arrow key is pressed. Now that the data is in, it can be analysed by selecting "analysis" from the menus. A number of options are available and most basic statistics (average, standard deviation etc.) are produced using the "basic stats" option. The best approach is to play around with the data set provided (Test.sts) and see what happens. IMPORT/EXPORT OF DATA: The easiest way to import/export data is to use a spreadshseet that supports space separated values (called CSV space in TurboCalc). Most versions of TurboCalc do this (3.5 and above definately - earlier ones possibly). Data can be imported from spreadsheets in the following way. In the first row of the spreadsheet put: CELL 1 CELL 2 CELL 3 Number of cases Number of variables Missing value In the next row of the spreadsheet put: CELL 1 CELL 2 CELL ..etc Variable name 1 Variable name 2 Variable name... The variable name must be a maximum of 8 characters. The data follows below the variable names. Statistics should be able to import such a file without problems, as long as you correctly specify the number of variables, cases and the missing value in the first row of the spreadsheet. Using the import function on TurboCalc you should be able to load in a Statistics file - again using the CSV space separated option. LIMITATIONS - READ CAREFULLY!! There are a number of waaknesses in this software that should be borne in mind before anyone starts using it for serious analysis. ~ The program will not accept negative numbers. ~ The paired statistics calculations work on the basis that both variables have the same number of cases - this limits the useulness of the F and independant T tests (hope to sort it our in the next revision). ~ The same missing value has to be used for every variable. ~ If you enter very large numbers i.e. above 9999,999.00 screen formatting can become a bit dodgy. Entering huge values (like population figures) is best done by inputting, e.g. 12.1 for 12.1 million etc.. Although you can enter values with more than two decimal places and they will be used in the calculations - they will not be displayed in the editor. ~ The latest non-parametric routines have not been extensively tested. The program opens on a hi-res interlace screen. If you have a 1084 monitor or a SCART T.V. you may find that setting your workbench screen to NTSC (Workbench 2.04+ only) makes the screen more readable as well as saving memory and reducing flicker. If you do this is combination with the MagicTV program you will very nearly have a flicker free interlaced display on a TV! To set up for NTSC move the NTSC monitor into your devs/monitor drawer. Add the line "sys:devs/monitors/NTSC" to your startup- sequence immediately before the line "LoadWB". Now double click on the NTSC picture in devs/monitors and go to prefs/screenmode. Select NTSC Hires laced and then save. Now reboot. You may have to adjust the workbench palette a bit to reduce flicker. As you would expect I accept no responsibility if it blows up your monitor (not all PAL TV's can handle NTSC - though Sony and Philips ones seem to be OK), but it works on mine! Very minimal graphics are included and at present results cannot be output to a printer. One very cludgy way around this is to use a screen dump program. You will make best use of the program if you have a compatible spreadsheet. Any graphs, barcharts etc can be produced using a spreadsheet. I see no point in spending hours duplicating the graphing facilities of spreadsheets. The software is reasonably crashproof (I have used it a lot with large data sets over the last ten years and during my Ph.D.). The production of the later versions has resulted in some substantial changes to code and some new bugs must have been created. Furthermore if you try silly things - you may be able to crash the program, so save your data regularly. If you register by email I can let you know about upgrades. If you want to moan: remember what you paid for this program and how much statistical software can cost! FUTURE: I hope to iron out any further bugs and add: better help and data labels. Development has been pretty slow lately as a result of the Amiga's troubles. I would prefer to develop in a more portable Basic dialect, so that I can compile it on other OSs - in case the Amiga goes totally belly up. THE AUTHOR - BNR CONSULTING Ltd: I work as a freelance market analyst and economist and have particular experience in the transport/travel/tourism field. Previous experience includes: ~ demand forecasting for Sydney airport rail link. ~ demand forecasting for Sydney LRT system. ~ customer satisfaction surveys for transport undertakings. ~ assiting in railway pricing research for UIC/CER ~ working on fourth framework EU projects. Depending on other commitments, I and my associates are available as contractors and to conduct research for clients in any part of the world - at very competitive rates. Particular specialties include: ~ Demand forecasting (prediction of consumer behaviour). ~ Designing, analysing and conducting market surveys. ~ Report writing and presentation. ~ Stated Preference (Conjoint analysis). ~ Public transport planning ~ Computer demand models (basic programming and spreadsheets). For further information contact me via email. HISTORY: VERSION 1.10 ENHANCEMENTS (1988): The program has been made more multi-tasking friendly and is less processor hungry when waiting for user input. There is a new title screen. Control of the display while viewing data has been improved. The amount of data that the module can handle has been increased from 300 to 500 values. A number of bugs have been removed from the regression module that could give rise to, 'division by zero errors'. The correlation matrix display has been enhanced. A disk swapping bug causing, 'unknown volume' error has been cured. Some preparations have been made to merge the two modules at a later date. The Regression predictions section now works correctly and is more closely linked to the generation of the regression equations. Finally it is easier to increase the number of cases in the data. VERSION 2.00 ENHANCEMENTS (1995): Program sets up its own hi-res interlace screen with flicker minimising palette and selects Topaz 11 font for a consistent appearance on all systems. Better presentation and more user friendly. Many algorithms speeded up. Improved error trapping and a number of bugs removed. Vastly improved data entry/editing. Pull-down menus. Ability to conduct analysis on single variables added. Independent t test, coefficient of variation, minimum, range and pooled standard deviation added. Speech removed, beeps added. Improved export of data. VERSION 2.10 ENHANCEMENTS (1991): Various bug fixes. ARP file requester added. Data recode added. Text input area cleared after error. Data transform debugged and moved to data entry module. VERSION 2.20 ENHANCEMENTS (1996): ASL file requestor added Fixed bug in add variables module. Added Cross-tabs. Improved User-interface. VERSION 2.30 ENHANCEMENTS (1997): Rewritten for Blitz Basic 2.10, significantly faster. VERSION 2.31 ENHANCEMENTS (Jan 1998): Improved import/export of data. Slight speed increase. VERSION 2.40 ENHANCEMENTS (April 1998) Program now checks for PAL/NTSC and opens window accordingly. Made it easier to exit program! Added in Wilcoxon, Mann-Whitney and Kruskall Wallis tests. More multitasking friendly. VERSION 2.41 ENHANCEMENTS (March 1999) Fixed a bug in median calculations.