Jump to content

Talk:SegReg

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Notability tag removed

[edit]

On 21 September 2016 User:Qwfp placed the following template in the article's heading:

See: [1]

It is so easy to throw tags around without verifying their justification.

A google search with key words: "segreg segmented regression" gave me on the first 10 pages 31 references to scientific articles and publications in international journals or university papers in which the SegReg software was actually used and applied. Hence, no lack of notability, I assume.

The 31 references were put in a section ==Bibliography==. I find this exaggerated, but what can you do before removing such threatening templates. Asitgoes (talk) 15:45, 8 October 2016 (UTC)[reply]

As the tag says, the independent sources need to "provide significant coverage of it beyond its mere trivial mention". No number of papers each saying something like "we peformed segmented regression using the program SegReg" can constitute significant coverage. Significant coverage would be a software review by an independent author in a reliable source, or a section discussing the sofware in a book by an independent author published by a reputable publisher. See Wikipedia:Notability (software). Qwfp (talk) 18:27, 8 October 2016 (UTC)[reply]
There are various references that say more than merely "we used SegReg". For example this article is reviewing: "The analysis of the long-term temporal trends was conducted by segmented (piecewise) linear regression using the SegReg free software program. (. . . . . .). The segmented regression approach introduces a breakpoint in the data, where appropriate, allowing for a broken or discontinuous line across all data points. In SegReg a total of 7 models can be fit to the data, for example, no breakpoint, two sloped lines or two parallel lines with different Y intercepts. The model of best fit, as well as the breakpoint, is selected by SegReg according to the amount of variance explained by the model and a significance test". This is not a trivial mention. Asitgoes (talk) 19:33, 11 October 2016 (UTC)[reply]
The bibliography was sensibly removed.
It may be opportune to add another reference reviewing SegReg and expressing a preference: "Segmented linear regression analysis [22] on visual acuity and tracing errors was performed in SegReg program. [23] This program performs fitting of several regression models to data and selects a model with the best fit based on the highest coefficient of explanation and on significance testing. If a break point, i.e. a sudden change in the relation between the predictor and the dependent variable, is present in the data, the program finds the location of the break point and provides regression functions for segments of the data before and after the break point. Other statistical analyses were performed in SPSS v.20 (IBM SPSS Statistics)."
There is also a paper where SegReg was tested: "In order to test for thresholds in d-a relationships, we employed segmented linear (piecewise) regression with a break-point (e.g., [39]). Analyses were conducted with segmented regression software [40], which introduces breakpoints and calculates separate linear regressions for each segment. The optimal breakpoint is that with the smallest confidence interval. The value of introducing breakpoints is evaluated by assessing whether the piecewise regression models perform significantly better than linear regression models. The program selects the function type that maximizes the coefficient of determination (R2), and passes a test of significance based on an alpha value of 0.05 (degrees of freedom are corrected down appropriately as number of parameters increases with the different models [41], [42]). In addition, we assessed whether model selection by these criteria was consistent with that employing the bias-corrected version of Akaike's information criterion."
In this publication an argument is given: "In response to the claims that logarithmic estimation profiles should more accurately be described as two separate (bi-) linear profiles, and actually reflect separable processes (Ebersbach et al., 2008), we subjected the estimation data to bi-linear regression analysis using SegReg software (Oosterbaan, 2002). SegReg carries out a segmented linear regression by finding a breakpoint (if possible) in order to fit two separate linear functions that maximise the explanatory power of the resulting statistical coefficients. One argument for bi-linearity in estimation profiles is that young children in particular hold linear mental representations for numbers they are familiar with, but they guess when estimating numbers above this range (e.g. Berteletti et al . , 2010) thereby producing the more or-less flat profile that has been taken by others to be part of a continuous logarithmic curve.
More reviews can be cited if required. Asitgoes (talk) 12:51, 19 October 2016 (UTC)[reply]