New England Users Group

Founded 2009 · Six states · Notice no. 63

Autumn 2026 · Worcester & Portsmouth · Prototype issue

Every meeting leaves behind something another member can check.

Membership is free. It is open to anyone in the six states working with measurement, quality or experimental data — including students, and including people who have never presented anything.

We are a volunteer group of anyone who works with data — statisticians, engineers, medical professionals, and researchers in every domain — who meet in labs, mill buildings and lecture halls from New Haven to Portland to work through real analyses — the messy ones, with the data attached.

A meeting is not finished when the slides end. It is finished when somebody can rerun it: a reproducible report, a scripted example, or a finding stated plainly enough that a stranger can disagree with it on the evidence. Sixty-three of those now sit in the index, and each one names the data behind it.

In the twelve months to August 2026 the group grew to 486 members , published 63 reports , met 18 times across 7 host sites in 5 states, and answered 91% of the questions put to it — median first reply, eighteen hours.

i Plate — the group as a surface Drawn from the membership roll, 6 Aug 2026

Kernel density over 486 member towns

Half the membership stands inside one contour, drawn around Boston, Worcester and Providence — and it pays no attention whatever to the state lines.

The relief below is not a photograph or a shaded map. Height at any point is the value of a probability density built from where members actually live: a sum of Gaussian kernels, one per town, drawn as west-to-east traces over the six state boundaries. Turn the plate to read the ridges against each other.

Isometric relief plot of member density across the six New England states.

A single ridge rises over eastern Massachusetts, peaking between Cambridge and Boston, with lower shoulders over Worcester, Providence, Hartford and the New Hampshire seacoast. Maine, Vermont and northern New Hampshire are nearly flat.

Density trace Contour enclosing half the members State boundary at zero

Drag the plate to turn it. With it focused, ← → rotate and ↑ ↓ tilt; Home restores the plate.
Adjust the surface

Each layer is an ordinary closed-form function of longitude x and latitude y; the plate draws their sum. The first is the figure as published; the other three are there for the pleasure of it — turn them on and the surface stops describing the membership and starts describing arithmetic. Longitudes are scaled by cos 43° throughout, so one unit means the same distance in both directions.

z(x, y) = f₁

Traces are cut where they leave the six states, so the silhouette you see is the regional boundary itself rather than a rectangular window. Nearer traces hide farther ones, which is what makes the ridge read as a solid.

Layer definitions are held as data, so the same four expressions can be handed to a Desmos 3D calculator through setExpression when a production API key is in place. Until then the plate is drawn locally on a canvas, with no third-party script.

Read the plate as a table
Member density at the twelve kernel centres, relative to the Cambridge–Boston peak (1.00). Members within roughly twenty-five miles of each town.
Town State Relative height Members
Cambridge & BostonMA1.00151
WorcesterMA0.6258
ProvidenceRI0.4844
Hartford & New HavenCT0.4441
NashuaNH0.3834
PortsmouthNH0.3429
PortlandME0.3026
BurlingtonVT0.2823
SpringfieldMA0.2219
GrotonCT0.1817
Fall RiverMA0.1615
Bangor & OronoME0.1412

Source · NEUG membership roll, 486 records, aggregated to twelve town clusters; exact addresses are never published. Method · Gaussian kernel density, bandwidth 0.40 degrees, evaluated on the drawing grid at each frame. Boundaries · public-domain US state geometry, thinned to 640 points. Compiled 6 August 2026.

ii Forthcoming gathering All five dates below

Seventy-one days out. Registration is open and the room seats sixty. Street parking is free after six; the Innovation Studio is a nine-minute walk from Union Station.

Saturday 17 October 2026 · 5:30 – 8:00 pm · Worcester, Massachusetts

Design of experiments in practice

At the WPI Innovation Studio, Room 108. Hosted by Nadia Okonjo of Chelmsford Precision Optics, who will walk a two-level screening design from the first sketch to the run sheet. Bring a design you are stuck on — the last forty-five minutes are working time, not questions.

Register for Worcester 42 of 60 seats claimed

iii Featured finding Measurement exchange · Quality engineering

Screening designs cut setup scrap by a third at four pilot sites. The comparison lines drifted the other way.

Four member plants adopted two-level screening designs for changeover setup in January 2026. Eleven comparable lines kept one-factor-at-a-time practice. Both groups are plotted at monthly resolution below; the marked point is the most recent observation.

The two groups were not randomised — plants chose to join. Read the gap as an association, not an effect. The capability report in the index below repeats the comparison with matched fixtures.

Monthly setup scrap rate, pilot sites versus comparison lines

Point at the plot, or tab into it, to read a month directly.

A line plot of two series across twelve months of 2026, with a data table below.
Read the figure as a table
Month Pilot sites Comparison Difference

Source · NEUG measurement exchange, fifteen lines, self-reported monthly totals. Method · two-level screening designs against one-factor-at-a-time. Compiled 6 August 2026 by the reports committee; the workbook is attached to the report.

iv Recent reports Complete index, 63 entries

Every entry states its finding in one sentence and ships with the data or the script behind it. Two members read each report before it is listed; the review notes stay attached.

  1. Process capability across three production lines

    Line 2 clears the 1.33 capability threshold only on first shift, and the gap tracks fixture age rather than operator. Recommends a fixture-hours covariate in the monthly review.

    M. Alvarez & R. Chen2 August 2026Quality engineeringData and script attached12 min

  2. Assay drift after a reagent lot change

    A single lot change explains 62% of between-run variance. The control chart flagged it four runs before the release review did — the chart was simply not being read.

    P. Nkemdirim24 July 2026Life sciencesData attached8 min

  3. When binning hides the shift you are looking for

    Three common bin widths applied to one Merrimack Valley dataset produce three different conclusions. Only one survives resampling; the other two are artefacts of the histogram.

    J. Whitcomb11 July 2026VisualizationMethod notes6 min

v Calendar October 2026 – January 2027

Meetings held by state, 2026 to date. Hosting rotates; two 2027 slots north of Boston are still open.
Massachusetts
Connecticut
New Hampshire
Rhode Island
Maine & Vermont
5 dates

  1. 17 Oct Sat

    Design of experiments in practice

    Worcester, MA · WPI Innovation Studio · 5:30 – 8:00 pm · workshop, bring data

    In person 42 of 60 Register

  2. 29 Oct Thu

    Scripting a reproducible monthly report

    Online · 12:00 – 1:00 pm · clinic format · captions provided

    Online 96 enrolled Save a seat

  3. 12 Nov Thu

    Measurement systems analysis for small labs

    Portsmouth, NH · 4:00 – 6:30 pm · also streamed, parking on Deer Street

    Hybrid 31 attending Register

  4. 4 Dec Fri

    Winter benchmark: variation in small-batch lines

    Providence, RI · Rising Sun Mills · 5:00 – 8:30 pm · dinner included

    In person Opens 20 Oct Add to calendar

  5. 15 Jan Fri

    Newcomers' hour: reading someone else's analysis

    Online · 9:00 – 10:00 am · recorded · no preparation expected

    Online Open enrolment Save a seat

vi Questions before the group Ordered by response health, not volume

A question is unanswered until a member marks a reply as the answer. Two of the five below are waiting on a small reproducible dataset rather than on advice — which is usually the faster thing to supply.

5 questions

  1. Interval boundaries when bins are built from a rounded columnneeds example

    S. Doiron · Visualization · 2 hours ago

    0 replies · 126 views

  2. Blocking a split-plot when the hard-to-change factor is the ovenopen

    R. Kalinowski, Nashua NH · Design of experiments · 9 hours ago

    4 replies · 318 views

  3. Fastest way to reshape forty monthly exports into one tall tableanswered

    T. Moreau, Burlington VT · Scripting · yesterday

    11 replies · 742 views

  4. Is a Gauge R&R defensible with only two appraisers?needs domain expert

    H. Barros, Groton CT · Quality engineering · 2 days ago

    1 reply · 205 views

  5. Reading a capability report someone else builtanswered

    E. Ferreira, Fall River MA · Getting started · 3 days ago

    7 replies · 489 views

Join, and bring one unfinished thing.

There is no fee and no application. Add your name to the list and you will get the month's notice, the reports as they are published, and the right to put a question in front of four hundred people who have probably met it before. What follows is what is actually useful this month.

  1. Read one report end to end. Start with the assay drift report above — the data and the script are both attached, so disagreeing with it is cheap.
  2. Answer a question tagged “needs example”. Two are open. Both want a small reproducible dataset rather than advice.
  3. Bring a stuck analysis to Worcester. Working sessions run for the last forty-five minutes of every in-person meeting.
  4. Host a 2027 meeting at your site. We need three hosts north of Boston. A room, a projector and coffee is enough.

Add your name to the list