From perlman@mail.cis.ohio-state.edu Wed Jul 24 10:44:40 1996 Return-Path: Received: from mailer.oclc.org (mailer.dev.oclc.org) by fssun09.dev.oclc.org (4.1/SMI-4.1) id AA11986; Wed, 24 Jul 96 10:44:39 EDT Received: from mail.cis.ohio-state.edu by mailer.oclc.org (5.0/SMI-4.1) id AA22212; Wed, 24 Jul 1996 10:44:38 +0500 Received: from colon.cis.ohio-state.edu (colon.cis.ohio-state.edu [164.107.17.3]) by mail.cis.ohio-state.edu (8.6.7/8.6.4) with ESMTP id KAA18442 for ; Wed, 24 Jul 1996 10:44:30 -0400 From: Gary Perlman Received: (perlman@localhost) by colon.cis.ohio-state.edu (8.6.7/8.6.4) id KAA16272 for perlman; Wed, 24 Jul 1996 10:44:30 -0400 Date: Wed, 24 Jul 1996 10:44:30 -0400 Message-Id: <199607241444.KAA16272@colon.cis.ohio-state.edu> To: perlman@cis.ohio-state.edu Subject: Notes/stat Content-Length: 6402 Status: O Sugggested changes for |STAT PC-SIG #s 990-992 @name1 name2 ... nameN use in regress anova pair contab etc (any parseline) also used as comment lines for easy filtering @labels: name1 name2 ... nameN in colex - output is automatic for fields (unless type checking or formatting is sued) to extract colums requires lookup by specol Only do analysis of named columns (e.g., in regress) 1. Use of column names in expressions dm s1 or x1 -> height width e.g., height*width -> x2*x3 colex height width height width -> 2 3 2. Outputting names linex - just echo, but don't count? colex - output, but truncate to fmt? dm - use ptress to produce new column label? See Mail/unixbox for background Modular design vs monolithic design how to avoid creeping featurism - POET Analyze WRT Norman's 7 stages of execution/evaluation A | B | C | D result can be probed at any point (using less or (ugh) intermediate files) remember the IO program? /n/ship/0/perlman/stat/Old/io.c How does this generalize to the good aspects of the UNIX philosophy Problems with |STAT survey little idea of background of respondent most respondednts will be UNIX users |STAT Survey The following survey is to help me answer a question about user interfaces that has been interesting me for several years: Given that many packages are available at low cost, especially to educational institutions, why do people continue to use |STAT programs? Despite having a command-line interface, minimal documentation, and missing functionality, |STAT seems to satisfy some needs that are not met by other packages that are available. It might be that |STAT can be copied freely, but my intuition is that there are some fundamental user interface issues involved, but I do not want to discuss them before I try to get some survey results. So, if you are a |STAT user, then please take a few minutes to answer the survey by placing numbers and text after the double colons. I am concerned about this appearing to be a market survey, which it is not. I will, in anti-marketing fashion say that I have no plans to change |STAT. |STAT will not suddenly become a commercial product but will continue to be distributed under the same conditions as it has for the past ten years. Anyone who wants the raw data or analyzed data form the survey can request that I email it to them. I reserve the right to remove identifying marks from the responses. For answers requiring an estimate of frequency, use the scale: 1 never 2 seldom 3 sometimes 4 often 5 all the time For answers requesting multiple answers, feel free to add extra lines. [1] How often do you use |STAT programs? :: [2] Which |STAT programs do you use most often for analyzing data? [2.1] ________ :: [2.2] ________ :: [2.3] ________ :: [2.4] ________ :: [2.5] ________ :: [3] How often do you use: [3.1] individual commands :: [3.2] pipelines of commands :: [3.3] scripts or batch files :: [4] What non|STAT programs do you use most often with |STAT? [4.1] ________ :: [4.2] ________ :: [4.3] ________ :: [4.4] ________ :: [4.5] ________ :: [5] What |STAT programs do you use outside of data analysis (e.g., formatting)? [5.1] ________ :: [5.2] ________ :: [5.3] ________ :: [5.4] ________ :: [5.5] ________ :: [6] What other data analysis packages do you have easy access to, and how often do you use them? [6.1] ________ :: [6.2] ________ :: [6.3] ________ :: [6.4] ________ :: [6.5] ________ :: [7] In decreasing order of importance, list 5 reasons why you use |STAT. [7.1] [7.2] [7.3] [7.4] [7.5] [8] In decreasing order of importance, list 5 ways that |STAT is easy to use. [8.1] [8.2] [8.3] [8.4] [8.5] [9] In decreasing order of importance, list 5 ways that |STAT is hard to use. [9.1] [9.2] [9.3] [9.4] [9.5] [10] What new functionality would you like to see in |STAT? [10.1] [10.2] [10.3] [10.4] [10.5] [11] Please write comments on the following areas about |STAT: [11.1] The functionality provided :: [11.2] The documentation (paper and online) :: [11.3] The input formats :: [11.4] The command line options :: [11.5] The output formats :: topic index for stat keyword section for manual entries :desc (runtime allocation) :desc incorporate stats printout options :is dm the only program using strings (yes) convert : anova: polynomial : number: length checks for loss of precision : contab: crosstab breakdown pair add way to add labels to points Possible, and relatively easy to add a label to the same line as the data pair, however, I am not sure how to handle something like: X Y A X Y B where X and Y are data pairs and A and B are labels. What would you plot? I guess something like a * could indicate a conflict of labels. :use NA for unknown values in programs : calc:UNDEFINED : incorporate in validata for checking : NA would cause removal of all related data : NA is an integer : # of NA data should be mentioned NA handling in DM : comment lines (could make own fgets, or make ncols == 0 (blank line)) :confidence intervals for desc -c and -i used (-i could be made -w) dsort could produce the same effect as Jay McClelland's dt in the event of a tie, data columns are averaged, listed, or summed a b x a c y a b z ============== a b f(x,z) a c y tie reoslution can be done on output anova onlint declared as int/void between UNIX and DOS linex print all but the lines named series add -n option to specify the number of elements in the series series 10 20 15 incr = (high - low) / n; for (i = 0; i < n; i++) element = low + incr * i; simple transformations (e.g., exp, log, geom) in dm? parseline could work by setting pointers to fields and not clobbering lines parseline: leading space can be part of a field, with no effect on atoi or atof (some effect on strings, especially if left justified) This would allow outputs of parselined inputs to match the inputs colspec needs ot be able to refer to Nth, N-1st column 1-N N-1 --> N..1 may need to change range character to : or .. Multivariate factor analysis discriminant analysis canonical correlation multidimensional scaling GLIM Comparisons confidence intervals contrasts polynomial decomposition Graphics Library ticklines label lines Storage compare time for calloc (HUGE) vs data[HUGE]