Xan Gregg
LightNews LightNews
banner
xangregg.bsky.social
Xan Gregg
@xangregg.bsky.social
1.7K followers 1.8K following 240 posts
Engineering Fellow at JMP, focused on #DataViz, preferring smoothers over fitted lines. Creator of JMP #GraphBuilder and #PackedBars chart type for high-cardinality Pareto data. #TieDye #LessIsMore
Posts Replies Media Videos
xangregg.bsky.social
Trying the same thing with JMP 19's constrained p-splines. Nicely similar to the R GAMs, but JMP's CIs are bootstrapped instead of analytical, which is why they're not symmetric. #dataviz
Scatter plot with two smoothers for a 32-row cars data set. X is HP and Y is displacement. One smoother is constrained to be monotonic and flattens out at the end where the other one dips down.
November 13, 2025 at 1:30 PM
xangregg.bsky.social
Did you know (robust) loess fitting can fail if the data is already smooth? I made a notebook that shows the flawed fit (red) along with a possible improvement (blue), using Cleveland's original demonstration curve. #stats
observablehq.com/@xangregg/lo...
Screenshot of an interactive Observable notebook. Includes a graph containing data points and curves. A gray curve is the function used to generate the data by adding a small amount of noise. A red curve is a robust loess fit. A blue curve is a robust loess fit using a new outlier rule.
November 6, 2025 at 5:50 PM
xangregg.bsky.social
New blog post looking at some recently-shared NCAA football player data. The scatterplot is percent drafted to NFL against average player high school rating by college. Also trying out inward-jittered, smoothed dot plots.
rawdatastudies.com/2025/10/26/n...
Scatter plot with a fitted regression spline of "average play rating" on X and "NFL draft rate" on the Y. Dot chart of player rating on the X axis and Draft status (true or false) on the Y axis. values are jittered inward in the Y direction, resulting in two smoothed Wilkinson-like hexagonal-grid dot plots with the top one inverted.
October 27, 2025 at 2:14 PM
xangregg.bsky.social
Dot plot #dataviz comparison: ratings of FIDE chess Grand Masters via Tidy Tuesday.
1 Nearest stacks (Wilkinson)
2 Smoothed stacks
3 Smoothed hexagonal grid
4 Exact position (beeswarm)
Smoothing trades delta-x for spikiness (deviation from kernel density estimate).
Dot plot of FIDE Grand Master chess ELO ratings. Most are between 2400 and 2600. Each dot is placed in the nearest stack. Source data: https://github.com/rfordatascience/tidytuesday/blob/main/data/2025/2025-09-23/readme.md Dot plot of FIDE Grand Master chess ELO ratings. Most are between 2400 and 2600. Each dot is placed in the nearest stack or the next nearest if it helps avoid spikes. Dot plot of FIDE Grand Master chess ELO ratings. Most are between 2400 and 2600. Each dot is placed in the nearest cell of a hexagonal grid, or a nearby cell to avoid spikes in the profile. Dot plot of FIDE Grand Master chess ELO ratings. Most are between 2400 and 2600. Each dot is placed at its true x position, adjusted vertically to avoid overlap.
September 27, 2025 at 12:17 PM
xangregg.bsky.social
JMP 19 is out (free trial available), and I wrote a blog post about the main things I worked on. Constrained smoothers, jitter options, easier arrows, parallel y axes, ... #dataviz
community.jmp.com/t5/JMPer-Cab...
Image from linked blog post showing hexagonal jitter example. Penguin body mass colored by sex.
September 19, 2025 at 12:53 PM
xangregg.bsky.social
‪Nice wxdu.org set tonight from @gavinyamey.bsky.social‬, especially the spoken word pieces.
wxdu.org
WXDU 88.7 FM Durham - Duke University Radio
wxdu.org
September 11, 2025 at 11:53 PM
xangregg.bsky.social
Yay, I was able to reproduce the lines in this chart precisely from the raw data. The original shows summary dots where mine shows raw data dots, and at a couple zoom levels. The power of statistics; signal and noise. www.nature.com/articles/s41...
Figure 2a from Nature paper https://www.nature.com/articles/s41586-025-09321-3. Shows a line chart over a bubble chart. Recreation of previous chart but showing raw data dots instead of summarized bubbles. Same as previous chart, but with the Y axis zoomed out a bit. Same as previous chart, but with the Y axis zoomed out to show all the data (and fitted lines look almost flat).
August 22, 2025 at 7:26 PM
xangregg.bsky.social
What to make of a paper that shares a ton of well-organized data and code for its charts, but not enough detail for analysis? PII concerns, maybe.
Curiously, these line charts are random data, suggesting steadier step counts. www.nature.com/articles/s41...
Excerpt from Figure 1 of https://www.nature.com/articles/s41586-025-09321-3, showing part of a US map with two inset line charts indicating the daily step counts before and after a move from Seattle to San Francisco, The lines are relatively flat, at around 6000 steps for Seattle and 6700 steps for San Francisco. Screenshot from a Python notebook shared with the paper. Code part reads: with plt.rc_context({'figure.autolayout': True}): fig, ax = plt.subplots(figsize=(4, 2)); pre_x = range(-35, -5); y = np.random.normal(loc=from_df.loc[from_df['from_loc'] == 'Seattle, WA', 'pre_avg'], scale=50., size=(len(pre_x), )); plt.plot(pre_x, y, lw=5., c='#aa3939'); plt.ylim(5800, 7000); ax.grid(False); for item in ([ax.xaxis.label, ax.yaxis.label] + ax.get_xticklabels() + ax.get_yticklabels()): item.set_fontsize(axis_fontsize); ax.set(xlabel=r'Days from Move $\left(t - t_{move}\right)$', ylabel='Daily Steps', xticks=range(-35, -4, 10)); fig.tight_layout() plt.savefig('../output/fig1b_subplots/seattle_from.png', dpi=600);
August 20, 2025 at 12:48 AM
xangregg.bsky.social
Quick dot plot #dataviz study with 2500 US city Walk Scores. Plain dot plot (exact because scores are integers), with smoothing (±1), and with hexagonal placement (±0.75). Data from www.walkscore.com
Dot plot of Walk Scores (0-100 scale) of 2500 US cities. The bulk is centered around 25-45 range. Occasional spikes with one big one at 37. Dot plot of Walk Scores (0-100 scale) of 2500 US cities. The bulk is centered around 25-45 range. Smoothed pile heights. Dot plot of Walk Scores (0-100 scale) of 2500 US cities. The bulk is centered around 25-45 range. Smoothed pile heights on a hexagonal grid.
August 17, 2025 at 4:57 PM
xangregg.bsky.social
A week ago I was crazy enough to email a paper's corresponding author for the data. No response.

"The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request."

Why is this data availability statement evil? A few reasons: ...
August 10, 2025 at 5:08 PM
xangregg.bsky.social
Rare sighting of letter-values plots in the wild. Nicely described in the caption as "plots which first identify the median, then extend boxes outward, each covering half of the remaining data." n=2.9M, so regular box plots would be swamped with outliers. #dataviz
arxiv.org/pdf/2402.14583
Chart with 5 letter-values plots, a variant of box plots.
August 1, 2025 at 12:13 PM
xangregg.bsky.social
Need to get the raw data behind a chart? Here's a walk-through of my PDF → SVG → CSV → Data techinque. #dataviz
rawdatastudies.com/2025/07/21/d...
Data extraction challenge – Raw Data Studies
Throughout my quests for raw data, I've learned a few techniques for find data lurking behind the charts. This walk-through shows a few of them,
rawdatastudies.com
July 21, 2025 at 2:32 PM
xangregg.bsky.social
The Secret of Data Science. I don't know if I'll ever get the chance to present this wisdom in public, so I'm sharing a rehearsal video from my rejected OutlierConf lightning talk submission. It really needs a live audience, though. #dataviz youtu.be/imRSlilIw5k
Secret of Data Science
YouTube video by The Graph Builder
youtu.be
July 11, 2025 at 5:20 PM
xangregg.bsky.social
I've written an explainer blog post for my "data strips" #dataviz workbench web app, trying alternative 1-D distribution summary plots. rawdatastudies.com/2025/07/05/d...
Data Strips Experiment
I built a “Data Strips” app to experiment with new ways of graphically summarizing the distribution of a single variable.. You can try it out or access the code on GitHub. This post wil…
rawdatastudies.com
July 6, 2025 at 3:52 PM
xangregg.bsky.social
Round 2 of my 1D #dataviz experiment at xangregg.github.io/data-strips/.
I realized my adaptive outlier idea was already done as Grubbs' test, which I've adapted for non-gaussian moments.
Added a couple thirds-based views. Here's 5000 random normal samples plus 2 outliers. The green ones use Grubbs.
14 one-dimensional views of a data set of 5000 random normal values and two larger outliers. Traditional HDR and Box Plot have outlier tests that don't scale this high, but Grubbs' test does.
June 30, 2025 at 6:55 PM
xangregg.bsky.social
6 old and 7 new 1D #dataviz. Trying Shortest Half with a twist: one break is allowed. Also half-sample mode & count-adaptive outlier thresholds. Here's 1000 random normal points plus two outliers. The green ones are new. Try it at xangregg.github.io/data-strips/
Screenshot of a web app displaying 13 one-dimensional horizontal views of the same data set. Summary of each: 1. Heatmap of fixed-sized intervals colored by relative count. 2. Kernel Density Estimation using a gaussian kernel and multiple bandwidths. 3. Rug plot with a vertical line at each data point, up to 2000 points. 4. Density Strip: KDE using color and a bandwidth multiplier of 50%. 5. Highest Density Regions plot with cut-off points at 0.0, 0.5, 0.95, and 0.99. 6. Highest Density Regions plot with cut-off points at 0.0, 0.5, and an adaptive outlier threshold based on a gaussian extrapolation of the middle region. 7. The shortest contiguous half that contains at least 50% of the data, applied iteratively to show 1/2, 1/4, 1/8, 1/16 and the shortest half mode. 8. Quantile regions at percentiles that correspond to equal intervals if the data is Gaussian. Each region would be one standard deviation wide. 9. Density Rug: Shortest regions of 20%, 50%, and 80% of the data, allowing for split regions penalized according to the Split Penalty. Other values are shown as a rug plot, except touching values are connected as a single region to avoid looking more dense than the shortest regions. 10. Shorth and Mode: Shortest half of the data as one or two contiguous intervals; any split is penalized according to the Split Penalty. A vertical line shows the "half sample mode" which is the iteratively applied shortest half. 11. Inner region shows Interquartile Range (IQR) of the data with a line at the median. 12. Inner region shows Interquartile Range (IQR) of the data with a line at the median. When the outer region extends beyond the common box plot whiskers (1.5 IQR), the endcaps are shown as arcs. 13. Box Plot: Inner region shows Interquartile Range (IQR) of the data with a line at the median.
June 23, 2025 at 5:02 PM
Reposted by Xan Gregg
sonjakuijpers.bsky.social
I am not at #Outlier2025 but I am sure there are many perspectives shared that I also care deeply about and I also want to point out that tomorrow I'll be giving insight into my personal thoughts on #dataviz at the Data+Women online data art episode 3.
usergroups.tableau.com/events/detai...
Data Art Episode 3 : Recalibrate | Tableau
Virtual Event - Data + Women Zurich
usergroups.tableau.com
June 11, 2025 at 3:13 PM
xangregg.bsky.social
This bilinear fit prompted me to try out a p-spline idea. P-splines are additive models that minimize adjacent slope differences. What if we relax the weight on the difference at the knot with highest curvature and re-optimize? ... Not bad.
Time series chart of day of peak cherry blossoming each year over time for over 1000 years. A p-spline smoother shows the overall trend, which sharply declines after 1890. Data from the parent post. A reference line at 1890 corresponds to the change point from the parent post's paper.
June 11, 2025 at 6:59 PM
xangregg.bsky.social
Adding to my collection of bold Data Availability statements. Malicious compliance, or am I missing something? alzres.biomedcentral.com/articles/10....
Picture of text from a paper reading: "Data handling and statistical analysis Data were analyzed blindly. Data are presented as the mean ± SEM. Statistical analysis was performed with SPSS 18.0 software. The test of Kolmogorov–Smirnov with the correction of Lilliefors was used to evaluate normal distribution and the test of Levene to evaluate the homogeneity of variance. Significance was analyzed by one-way ANOVA and Bonferroni’s multiple comparison post hoc test. Significance was considered when p < 0.05. Data availability No datasets were generated or analysed during the current study."
June 9, 2025 at 3:30 PM
xangregg.bsky.social
Neil Sloane on Numberphile never disappoints. Founder of oeis.org Online Encyclopedia of Integer Sequences.
youtu.be/pAMgUB51XZA
Amazing Graphs - Numberphile
YouTube video by Numberphile
youtu.be
May 26, 2025 at 9:06 PM
xangregg.bsky.social
The median lines for the two gray swarms look wrong, possibly swapped. And a couple outliers are dropped from the third group. Fig 3d from www.nature.com/articles/s41...
Chart of dot swarms for three x values and a log-scale response. Median lines are provided for each group. Remade chart of dot swarms for three x values and a log-scale response. Median lines are provided for each group. Dots are aligned to a grid.
May 26, 2025 at 8:07 PM
xangregg.bsky.social
Remaking radar charts; nine planetary control variables over time (past and future-modeled). Thought I could get by with all lines the same color and style but did make the two fast climbers thicker. #dataviz
Original www.nature.com/articles/s41....
Panel of four radar charts with 9 measures each. Original caption: "a–d, Control variables for planetary boundaries for 1970 (a), 2015 (b), SSP2 2030 (c) and SSP2 2050 (d). The green zone is the safe operating space, the pale orange represents the zone of uncertainty (increasing risk), and the dark orange is the high-risk zone. The planetary boundary itself lies at the intersection of the green and pale orange zones. The control variables have been normalized to the reference value, planetary boundary and the upper end of the uncertainty zone, as explained in the Methods. The control variables and values used are indicated in Table 1. The planetary boundary figures are a useful conceptual visualization tool, but still subject to limitations, as discussed in the literature." Source: https://www.nature.com/articles/s41586-025-08928-w Time chart with nine lines, each a separate planetary control variable, from 1975 to 2100. Values are normalized to a planetary boundary at 1.0. Area under the boundary is colored green and higher regions are yellow and orange.
May 26, 2025 at 2:32 PM
xangregg.bsky.social
Remaking a line chart as data points and smoother (and with a zero baseline). #dataviz
Line chart of US bicyclists deaths by year. From https://bikeleague.org/another-year-of-devastating-and-preventable-bicyclist-deaths/. Smooth trend line of bicyclists deaths by year with dots showing actual values. Original from https://bikeleague.org/another-year-of-devastating-and-preventable-bicyclist-deaths/
May 22, 2025 at 1:12 PM
xangregg.bsky.social
Another experiment replacing summary bars of integer data with jittered dots having a violin profile. Original #dataviz from link.springer.com/article/10.1...
Bar chart with three bars (control, <1h, >1h) and a truncated 0-10 response scale. Error bars and significance brackets are shown. Dor plot for three variables and 0-10 integer response values. Dots are jittered like a symmetric smoothed dot plot and create a violin plot look.
May 18, 2025 at 8:58 PM
Reposted by Xan Gregg
rsimmon.bsky.social
One thing I’ve noticed as people move away from the rainbow palette is more use of divergent palettes for sequential data. That’s not *always* a bad idea, but often is. This example inadvertently de-emphasizes states (Colorado) where people hear about climate change more frequently.

#dataviz 📊
A map of the US showing how most people aren't talking about climate change. Source: https://climatecommunication.yale.edu/visualizations-data/ycom-us/ The palette goes from dark red (100% of people in that state hear about climate change weekly), to a transition between pale orange and pale blue at 50%, to dark blue (people never hear climate change messages). In the U.S. no state is above 50%, so states with more frequent climate communication (Colorado, Massachusetts) are lighter than states with less (Oklahoma, West Virginia). This visually de-emphasizes those states.
May 16, 2025 at 6:50 PM