Editing Talk:2023: Y-Axis

<!--Please sign your posts with ~~~~ and don't delete this text. New comments should be added at the bottom.-->

"There are four kinds of lies: lies, damned lies, graphs, and statistics." [[User:Andyd273|Andyd273]] ([[User talk:Andyd273|talk]]) 13:37, 23 July 2018 (UTC)


To me this graph stands out as having something very wrong far more than those that limit the y axis to a short range. If the grid lines were several shades lighter however...  [[User:PotatoGod|PotatoGod]] ([[User talk:PotatoGod|talk]]) 15:44, 23 July 2018 (UTC)

Also I wonder if anyone can find a legitimate (non-misleading) use for the semi-semi-log plot? I’m sure there’s some scenario where it could be useful. Perhaps showing the population growth of a species, then when the growth levels out at the maximum sustainable level for its environment (I forget the proper term from high school biology) showing more detail of the small population changes or something like that? [[User:PotatoGod|PotatoGod]] ([[User talk:PotatoGod|talk]]) 15:52, 23 July 2018 (UTC)
:Frankly, it would be better to just use 2 separate graphs. Even if you explain to the reader that the scale changes mid-way, it would still be misleading on the subconscious level. The whole point of visualization is to allow the reader to utilize that sweet auto-processing power of our brains so that we don't have to think about what we are looking at too much. [[User:Jaalenja|Jaalenja]] ([[User talk:Jaalenja|talk]]) 17:59, 23 July 2018 (UTC)


Are there any IRL examples of this type of plot trick? I've never seen it

At first I thought the X-axis was logarithmic, because it lacks labels. This can also cause the sudden data jump.


There are no Y-axis labels and values, the x-axis dates are questionable, and the data points are even more questionable, resembling linear growth at really convenient spots. [https://amp.businessinsider.com/images/50b62c2669beddc340000005-320-185.jpg Fox News misleading graph]
@@ Line 2: / Line 2: @@
 "There are four kinds of lies: lies, damned lies, graphs, and statistics." [[User:Andyd273|Andyd273]] ([[User talk:Andyd273|talk]]) 13:37, 23 July 2018 (UTC)
-:Lies by omission!  ...not very funny, though[[Special:Contributions/162.158.106.66|162.158.106.66]] 13:50, 25 July 2018 (UTC)
 To me this graph stands out as having something very wrong far more than those that limit the y axis to a short range. If the grid lines were several shades lighter however...  [[User:PotatoGod|PotatoGod]] ([[User talk:PotatoGod|talk]]) 15:44, 23 July 2018 (UTC)
@@ Line 8: / Line 8: @@
 Also I wonder if anyone can find a legitimate (non-misleading) use for the semi-semi-log plot? I’m sure there’s some scenario where it could be useful. Perhaps showing the population growth of a species, then when the growth levels out at the maximum sustainable level for its environment (I forget the proper term from high school biology) showing more detail of the small population changes or something like that? [[User:PotatoGod|PotatoGod]] ([[User talk:PotatoGod|talk]]) 15:52, 23 July 2018 (UTC)
 :Frankly, it would be better to just use 2 separate graphs. Even if you explain to the reader that the scale changes mid-way, it would still be misleading on the subconscious level. The whole point of visualization is to allow the reader to utilize that sweet auto-processing power of our brains so that we don't have to think about what we are looking at too much. [[User:Jaalenja|Jaalenja]] ([[User talk:Jaalenja|talk]]) 17:59, 23 July 2018 (UTC)
-:Yes, specifically in anomaly or outlier detection before doing any feature scaling/normalization, regression, sampling, replace of missing values. For data modeling, Semi-log can help you detect if outliers affect your model or if your p-hacking based on outliers.  For a given programming language or software, semi-log plot has had their place when you were not able to do quantile-quantile plot, heteroskedasticity plots, etc.  In layman's terms, it can be beneficial to compare both the semi-log and non-logarithmic pot simultaneously to see how removing outliers or large value might change the plot or results.  However, there now are easily accessible specific heteroskedasticity and outlier functions in R and cookbooks in python that would allow you test for outliers and data dredging more rigorously than semilog plots. Therefore, semi-log plots for outlier/anomaly detection may be going out of style.  I am not sure if there are any science's that still rely on semilog plots in data exploration step of science.  Does anyone know of any applications of semilog plots are still used for a specific science today? --[[Special:Contributions/162.158.186.36|162.158.186.36]] 22:51, 24 July 2018 (UTC)
-:I would use semi-semi-log plot to compare exponential behavior of one dataset with linear behavior of another, but this would not be the intention of the comic because the two axes would be used for distinct datasets. [[Special:Contributions/162.158.63.118|162.158.63.118]] 14:34, 25 July 2018 (UTC)
 Are there any IRL examples of this type of plot trick? I've never seen it
-At first, I thought the X-axis was logarithmic, because it lacks labels. This can also cause the sudden data jump.
+At first I thought the X-axis was logarithmic, because it lacks labels. This can also cause the sudden data jump.
 There are no Y-axis labels and values, the x-axis dates are questionable, and the data points are even more questionable, resembling linear growth at really convenient spots. [https://amp.businessinsider.com/images/50b62c2669beddc340000005-320-185.jpg Fox News misleading graph]
-:I think you were onto something about the X-axis being logarithmic.  X-axis AND Y-axis are both logarithmic.  The trick is to realize that the X-axis is reversed.  The Y-axis is logarithmic between 50% and 100%, but the X-axis is logarithmic on the LEFT and AFTER the first tick mark. A readable symlog or x-axis semi-log plot has the logarithmic on the LEFT or AFTER the first tick mark.  This I think really highlights an important point that Randall is making with this comic: '''Whether you exaggerate tick marks to the range to data or adjust ticks to a range outside of the data, you ultimately skew the meaning of the plot'''  Both Y-axis trick and log-scaling are bad. --[[Special:Contributions/162.158.186.36|162.158.186.36]] 22:51, 24 July 2018 (UTC)
-:Yes, there is a programming example in python besides the Fox News one shown above.  You can reproduce this plot using the symlog function in python.  This is my first time posting in this wiki, so I am not sure if I should edit the page to include this example.  Here is a link: https://matplotlib.org/gallery/scales/symlog_demo.html .  Specifically, double symlog plot has a similar axis to Randall's picture.  You might notice that you can also do this R; however, it is intentionally much harder to do because of the very point Randall is making. --[[Special:Contributions/162.158.186.36|162.158.186.36]] 22:51, 24 July 2018 (UTC)
-:There is an interesting color version of the point Randall is making that was published today in livescience: [https://www.livescience.com/63153-brain-color-distortion-maps.html].  Turns out our eyes for color expect this kind of scaling distortion. --[[Special:Contributions/162.158.186.36|162.158.186.36]] 22:51, 24 July 2018 (UTC)
-:There is also a related problem for the case of discrete plots like bar charts called Waterfall charts. Waterfall charts are so bad, that their is saying in business, "Waterfall charts are how you lie to stakeholders".  Here is a deeper explanation: https://zebrabi.com/excel-waterfall-chart/ --[[Special:Contributions/162.158.186.36|162.158.186.36]] 22:51, 24 July 2018 (UTC)
-Here is an example of a peer-reviewed scientific paper using a mixed linear/logarithmic scale on both axes: http://dx.doi.org/10.1029/2004JA010829 (Figure 9, page 8) [[Special:Contributions/162.158.222.52|162.158.222.52]] 12:17, 30 July 2018 (UTC)