Home / Statistical Tools / Control Charts / EWMA / Options
Options¶
Want to build one first?
How-To works through a finished chart step by step, from sample data you copy off the page straight into Excel. The data has a point that is out of control, so the result has something to read rather than twenty points in a row.
How-To with GroupBy does the same thing with a GroupBy column, so one run produces a separate chart per group.
This page covers what is specific to the EWMA Chart dialog. The parts shared by every control chart, the Data Selection window, the Data tab, and how subgroups are formed, are described once on Create Control Charts. The pane that opens beside the finished chart has its own page, Task Pane. What each option means statistically is on the Math Details page.
The Dialog¶
There is no variant selector above the tabs. The ribbon has one EWMA entry, the dialog draws one panel, and the dialog title is EWMA Chart.
EWMA has three tabs, not four: Data, Limits & Estimation, and Rules & Display. There is no Capability tab, and its absence is not cosmetic. The engine switches the capability index and the histogram off for this chart, so no capability output is produced on an EWMA sheet under any setting.
Data Tab¶
The slot lists are Measurement Columns, X-Axis Labels (Optional), and the Subgroups group with its subgroup identifier drop-down. See Create Control Charts for what each one does.
Three things are specific to EWMA:
- Under Constant size: the lowest
nthe box accepts is 1, not the 2 the subgroup charts require. A subgroup size of 1 is a legitimate EWMA chart: each row is one value and the chart smooths the values themselves. The largest is 100, which is as far as the Shewhart constant tables reach. A value outside that range is refused with a message box reading "The subgroup size must be a whole number from 1 to 100.", and the box shows the value it held before. - A subgroup size of 1 changes the Limits & Estimation tab while you watch. The Estimate Sigma group is driven by the subgroup form, so setting the size to 1 replaces the three subgroup estimators with the three individual-observation estimators and adds the moving range length box and the Nelson checkbox. Setting the size back to 2 or more swaps them back. Nothing announces the change.
- Subgroup ID column: and Subgroups across columns can both produce subgroups of unequal size, because an identifier can repeat a different number of times and a row can have blanks in some of the checked columns. Unequal sizes are supported, and each subgroup's own size is used at its own point when the limits are computed.
Limits & Estimation Tab¶
This tab carries seven groups: Estimate Sigma, Transformation, EWMA Weight, and Historical Values on the left, and Control Limits, Estimate Limits From, and Split Control Limits on the right.
Control limits¶
The Control Limits group on the Limits & Estimation tab sets how the limits on the panel are obtained. On EWMA it offers two choices, not the usual three.
| Control | Type | Default | What it does |
|---|---|---|---|
| Shewhart | radio button | selected | The center line and the widening limits are computed from the data. |
| None | radio button | not selected | No limits are drawn. The points, the center line, the process statistics, and the outlier markers are all still produced. |
Manual limits are not offered on this chart. There is no Manual radio and no manual limits grid, because the limits are not a single band: they change at every point. The engine refuses manual limits on a time-weighted chart as well, so a Manual selection restored from a saved analysis is turned back into Shewhart when the dialog opens.
Choosing None also turns the out-of-control test off. Blanking the limits blanks the test result with them, so no point is flagged.
Math: Control Limits and Zones.
Number of standard deviations (k)¶
| Control | Type | Default | Accepts |
|---|---|---|---|
| Number of standard deviations (k): | text box | 3 | any number greater than zero, with no upper bound |
This is how far the limits sit from the center line, measured in estimated standard deviations. Lower it and the chart flags more points; raise it and it flags fewer.
Editing k also changes the first out-of-control test. That test asks whether a point lies beyond k standard deviations from the center line, and it reads this same box. That is why it is the one test with no parameter of its own: k is its parameter.
It decides whether there are zone lines at all. Zone lines are drawn only at \(k = 3\), where thirds of the band are one and two estimated standard deviations exactly. Set k to anything else and no zone lines are drawn on any panel, the Show zone lines check box greys with a notice saying why, and the four tests that read the zones do not run.
Where it is hidden. CUSUM has no k box. Its decision interval takes the place of a sigma multiplier.
Validation: blank, non-numeric, zero, and negative entries are all refused, with a message box:
The number of standard deviations must be a number greater than 0.
The box then shows the value it held before. Earlier versions discarded the entry in silence, so it was easy to believe a number had been accepted when it had not.
Math: Control Limits and Zones.
On EWMA the multiplier is applied to the standard deviation of the smoothed statistic at each point, so raising or lowering it scales the whole widening band, funnel and all.
Historical (known) values¶
The Historical Values group lets you supply a number instead of estimating it, so the limits describe a process you have already characterized rather than the data in front of you. Every box is a text box, every box starts blank, and blank means estimate it from the data.
| Control | Type | Default | What it replaces |
|---|---|---|---|
| Mean: | text box | blank | the mean behind the center line |
| Standard deviation (overall): | text box | blank | the overall sigma |
| Standard deviation (within): | text box | blank | the within-subgroup sigma |
| Mean (between): | text box | blank | the mean of the subgroup means |
| Standard deviation (between): | text box | blank | the between-subgroup sigma |
Which boxes appear depends on the chart, because a box for a quantity the chart never uses would be misleading.
| Chart | Boxes shown |
|---|---|
| Xbar-R, Xbar-S, I-MR | Mean, overall, within |
| I-MR-R/S | all five |
| EWMA, CUSUM | Mean, within |
| Levey-Jennings | Mean, overall |
Behavior worth knowing:
- A supplied value wins, and it is applied last. The estimate is computed from the data first and then overwritten, so a First N baseline still selects which points fed the discarded estimate, and still governs anything you did not supply.
- One number can change two. Supplying a within sigma also redefines the between sigma, and supplying a between sigma redefines the sigma of the subgroup means, because the remaining quantity is back-solved from the data.
- A supplied sigma carries no degrees of freedom, so it gets no confidence interval in the capability output. A supplied mean affects no degrees of freedom.
- One value covers every phase. With split control limits on, the same supplied number applies to all phases; there is no per-phase historical value.
- A transformation discards them for that run. Turning on Box-Cox or Johnson greys out the whole group and the values are not used. What you typed reappears if you turn the transformation off.
Validation: every one of these boxes is now checked in the dialog, and refuses with a message box naming the box. The three standard deviation boxes take a number greater than zero, or blank:
The historical standard deviation (within) must be a number greater than 0, or blank.
The two mean boxes take any number, or blank, because a mean may legitimately be negative:
The historical mean must be a number, or blank.
In both cases the box shows the value it held before. Earlier versions range-checked none of these boxes, so an unusable value passed the dialog and showed up as an unusable chart.
Math: Process Sigma Estimation.
EWMA shows two of those boxes, Mean: and Standard deviation (within):. The overall sigma box is not shown, because nothing on this chart is built from the overall estimate.
What Estimate Limits From (Baseline Estimation) does¶
It changes what the limits are estimated from. It never changes what is plotted. Every point always appears on the chart. Asking for the first 25 subgroups does not hide subgroup 26 onward. It means the center line and the control limits are computed from the first 25 subgroups and then drawn straight across the whole chart, so the later points are judged against the earlier baseline. That is the entire purpose of the setting: it is how you freeze a known-good period and watch what happens afterwards.
The same setting carries two labels. It is Estimate Limits From on the variable charts and Baseline Estimation on the attribute and rare-event charts, because those dialogs were written separately. The intent is identical, but the smallest count you may enter differs, and so do a few of the edge cases, so use the block written for this chart.
| Estimated from the selected points only | Computed from all the points regardless |
|---|---|
| The mean behind the center line | Every plotted value |
| The overall sigma | Each subgroup's own mean, range, and standard deviation |
| The within-subgroup sigma and everything it is built from | The minimum, the maximum, and the count |
| The between-subgroup sigma | The observed parts per million outside the specification |
| The moving-range estimates on an individuals chart | The histogram and the distribution fit |
| The estimated proportion or rate on an attribute chart | The out-of-control tests, which are applied to every point |
| The reported degrees of freedom | |
| The transformation itself |
Two of those rows surprise people:
- The transformation is fitted on the baseline too. A Box-Cox lambda or a set of Johnson parameters is fitted from the baseline points only and then applied to every row in the data. Change the baseline and you change the transformation, not just the limits.
- The degrees of freedom come from the baseline, so a short baseline widens the capability confidence intervals as well as changing the limits.
With split control limits on there is still only one count. It is reused inside every phase, capped at the length of that phase, so a phase shorter than the count simply uses all of its own points. The count never reaches across a phase boundary, and there is no way to give one phase a different baseline from another.
Math: Baseline and Limit Estimation.
Estimate Limits From¶
| Control | Type | Default | Accepts |
|---|---|---|---|
| All subgroups | radio button | selected | |
| First N subgroups: | radio button | not selected | |
| the count box beside it | text box | blank | a whole number of 5 or more. Its tooltip reads "Minimum 5 subgroups". |
The count is in plotted points, not in worksheet rows. On an individuals chart it counts points, and the moving ranges available inside that window are one fewer than the count at the default moving range length, fewer still if there are gaps.
| What you do | What happens |
|---|---|
| Select First N subgroups: and leave the box blank | The chart does not redraw and Finish switches off, with the reason beside it. No message box opens. See below. |
| Type a number below 5 | Refused with a message box reading "The first N subgroups must be a whole number of 5 or more.", and the box goes back to what it held before. |
| Type something that is not a number | Refused with "The first N subgroups must be a number.", and the box goes back to what it held before. |
| Type a number larger than the data | Every point is used. |
| Leave a blank subgroup inside the window | It does not use up one of the N. The window reaches further into the data to find real subgroups. |
| Exclude a point as an outlier inside the window | The same: it does not use up one of the N. |
The minimum of 5 is the variable chart engine's own: it refuses to compute limits from fewer than five points. The attribute and rare-event charts use a minimum of 4, and the T-squared and Generalized Variance chart uses 2.
Selecting First N subgroups with the box beside it empty stops the redraw, and says so quietly. Choose the First N subgroups radio before typing a count and the chart does not redraw, the reason appears beside the Finish button, and Finish switches off:
First N subgroups is selected, so the box beside it needs a number.
No message box opens for this one. You have selected the radio and have not yet had a chance to type, so the dialog waits rather than interrupting. Type a count and everything continues. A box you have not finished filling in is the one kind of refusal that stays quiet; a rule about two settings that disagree with each other still opens a message box.
The box is not filled in for you either. A count equal to every subgroup would mean the same thing as All subgroups, and it would write a number into the worksheet's settings note that nobody typed.
Before this, selecting the radio with an empty box redrew the chart from all subgroups while the settings note read "First 0 subgroups", so the note stated a baseline the chart had not used. A worksheet saved before this rule existed can still hold that state, and the settings note now prints All subgroups for a count of 0 rather than "First 0 subgroups".
A specification limit switches this off. Type a lower or an upper specification limit on the Capability tab and First N subgroups: greys out and the selection reverts to All subgroups, because a frozen baseline and a capability calculation disagree about which data describes the process. Levey-Jennings, EWMA, CUSUM and Z-MR have no Capability tab, so this cannot arise on them. There is no message, and clearing the specification limit makes the radio available again.
A target on its own does not switch it off, and the combination then fails
The dialog gate reads the LSL: and USL: boxes only, so a Target: with both specification boxes empty leaves First N subgroups: selectable. It should not be. A target on its own is enough to turn the capability calculation on, and the engine refuses to combine any capability calculation with a First N baseline, so pressing Finish in that state produces no chart at all and the message "Capability cannot be combined with a First-N baseline."
To avoid it, either clear the Target: box or set Estimate Limits From back to All subgroups.
Every phase still needs at least five usable points. With split control limits on, one phase too short fails the whole analysis with a message instead of a chart, whatever this setting says.
Validation: the count box refuses a number it cannot use, with a message box, and keeps what it held before. Earlier versions corrected the count in place and ran the analysis without reporting anything.
Two notes for this chart. The specification-limit gate described above cannot arise here, because EWMA has no Capability tab to enter a specification limit on. And with a constant subgroup size of 1 the labels still read All subgroups and First N subgroups: even though the chart is plotting individual values; read "subgroup" as "plotted point" in that case.
Estimate Sigma, subgrouped data¶
The Estimate Sigma group chooses how the process standard deviation behind the limits is estimated. The estimates differ in how they weight the subgroups and in how much they are disturbed by an unusual one, not in what they are trying to measure.
| Control | Type | Default | What it uses |
|---|---|---|---|
| Rbar (average range) | radio button | selected when it is the first method offered | the ranges of the subgroups |
| Sbar (average standard deviation) | radio button | selected when it is the first method offered | the standard deviations of the subgroups |
| Pooled standard deviation | radio button | not selected | all the within-subgroup variation pooled together |
Each dialog shows only the methods that suit its chart, and the first one shown is the one selected when the dialog opens.
| Chart | Methods offered | Default |
|---|---|---|
| Xbar-R | Rbar, Pooled standard deviation | Rbar |
| Xbar-S | Sbar, Pooled standard deviation | Sbar |
| I-MR-R/S | Rbar, Sbar, Pooled standard deviation | Rbar |
| EWMA, CUSUM | all three, when the data is subgrouped | Rbar |
If the data turns out not to be subgrouped, for example because a subgroup identifier column has a different value on every row, no subgroup method can be evaluated and a moving-range estimate is used instead, whatever the radio buttons show. No message is given.
Validation: this is a closed set of radio buttons, so there is nothing to validate and no message.
Math: Process Sigma Estimation.
With subgroups of two or more EWMA offers all three: Rbar (average range), selected when the dialog opens, Sbar (average standard deviation), and Pooled standard deviation. It is one of the few charts that offers the full set, because it has no dispersion panel that would tie it to one of them.
Estimate Sigma, individual observations¶
With one observation per point there is no within-subgroup spread to work with, so the estimate comes from how much consecutive points differ.
| Control | Type | Default | What it uses |
|---|---|---|---|
| Average moving range | radio button | selected | the mean of the moving ranges |
| Median moving range | radio button | not selected | the median of the moving ranges, which is much less disturbed by one large jump |
| Square root of MSSD | radio button | not selected, and shown only on the charts that offer it | the mean of the squared successive differences |
| Moving range length: | text box | 2 | a whole number from 2 to 100 |
| Use Nelson moving-range correction | check box | unchecked | on or off |
Moving range length is how many consecutive points each moving range spans. It disappears when Square root of MSSD is selected, because that method always uses consecutive pairs and would ignore it. It is shown only on the dialogs that offer the individuals sigma methods; a subgrouped chart that uses a moving range internally does not expose it, and its length is fixed at 2.
Use Nelson moving-range correction repeatedly removes unusually large moving ranges before averaging them, so that a genuine upset does not inflate the limits that are meant to detect it. It is available only with Average moving range: choose another method and the tick is cleared without a message, and the checkbox greys out. It considers only the moving ranges inside the baseline window, so a First N baseline restricts it further.
Validation: a moving range length below 2 or above 100 is refused with a message box, and the box goes back to the value it held before:
The moving range length must be a whole number from 2 to 100.
Earlier versions replaced such an entry with the nearest allowed value and showed the corrected number with no message.
Math: Process Sigma Estimation and Moving Ranges.
With a constant subgroup size of 1 that is the group EWMA shows, and all three methods are offered: Average moving range, selected when the group appears, Median moving range, and Square root of MSSD.
Where the sigma choice changes more than the sigma¶
Three of these interactions change the dialog or the chart itself, not only the number behind the limits.
- A subgroup size of one swaps the whole group. On EWMA and CUSUM, setting a constant subgroup size of 1 on the Data tab replaces the three subgroup methods with the three individual-observation methods and adds the moving range length and the Nelson checkbox. Setting it back to 2 or more swaps them back. Nothing warns you that the group you were looking at has been replaced.
- On I-MR-R/S the sigma choice decides which chart the third panel is. Choosing Rbar or the pooled standard deviation draws a range chart there; choosing Sbar draws a standard deviation chart instead. The row name in the manual limits grid does not change with it, so the grid still names both.
- Levey-Jennings has no Estimate Sigma group at all. Its limits come from the overall standard deviation of the data, which is the main thing separating it from an Individuals chart, so there is no method to choose and the unbiased-constants checkboxes are not shown either.
Math: Process Sigma Estimation.
The unbiased-constants checkboxes¶
Every estimate of a standard deviation from small samples runs slightly low on average. These two checkboxes decide whether that small bias is corrected. They have different defaults, which is easy to miss.
| Control | Type | Default | What it changes |
|---|---|---|---|
| Use unbiased constants (within sigma) | check box | checked | the correction applied to the within-subgroup estimate |
| Use unbiased constants (overall sigma) | check box | unchecked | the correction applied to the overall estimate |
The within checkbox greys out for some methods, because it would do nothing. It has an effect only with Sbar, with the Pooled standard deviation, and with the Square root of MSSD. With Rbar, the Average moving range, and the Median moving range the correction is always part of the estimate, so the dialog disables the checkbox rather than pretending the choice exists.
The overall checkbox is never greyed out, and it is a small correction because it is applied over the whole data set. It never moves a drawn control limit on any chart: the one chart whose limits come from the overall estimate is Levey-Jennings, and that chart shows neither checkbox.
Levey-Jennings does not show either box, and ignores both. Its overall standard deviation is always the uncorrected form.
Validation: checkboxes cannot be invalid, so there is no message. When a correction factor does not exist for the size it would be evaluated at, the correction is skipped and the uncorrected value is reported rather than nothing.
Math: Process Sigma Estimation and Control Chart Constants.
On EWMA the within checkbox is disabled under Rbar, which is the default with subgrouped data, and under the average and median moving range, which are the individuals defaults. It becomes live under Sbar, the pooled standard deviation, and the square root of MSSD, and there it changes the estimate itself, which changes the width of the whole band. The overall checkbox stays live but has nothing to move on this chart, whose limits never read the overall sigma.
EWMA Weight¶
This group has no counterpart on any other chart. It holds the smoothing weight, the one setting that decides how much of the chart's history each point carries.
| Control | Type | Default | Accepts |
|---|---|---|---|
| EWMA weight (λ): | text box | 0.2 | a number greater than 0 and up to and including 1 |
An italic line under the box reads "Between 0 and 1; default 0.2".
- What it does. Each plotted point is λ of the current value plus 1 minus λ of the point before it. At λ = 1 the chart plots the values themselves and the limits stop widening after the first point. As λ falls, each point holds on to more of the chart's history, the plotted line gets smoother and slower, and the band it is judged against gets narrower.
- It also changes the limits, not just the line. The width of the band at each point is built from λ, so changing the weight redraws the funnel as well as the plotted series.
- Validation: the box is read when it loses focus. A blank, a non-numeric entry, a value of zero or below, or a value above 1 is refused with a message box reading "The EWMA weight must be a number greater than 0 and at most 1.", and the box shows the value it held before. A value above 1 would make the recursion oscillate, which is why the upper end is closed at exactly 1. Earlier versions discarded the entry in silence.
Math: EWMA Math Details.
Transformation¶
A transformation replaces each measurement with a rescaled version of itself, chosen so the rescaled values sit closer to a normal distribution. Control charts assume a normal distribution and never fit any other one, so a transformation is the only remedy the product offers for skewed measurement data on a control chart.
| Charts | Transformation group box |
|---|---|
| Xbar-R, Xbar-S, I-MR, I-MR-R/S, EWMA, CUSUM | Offered |
| Levey-Jennings, Z-MR | Not offered. There is no control, and no transformation is applied. |
| p, np, Laney P', c, u, Laney U', G, T | Not offered. These charts plot counts, proportions, or waiting times and have their own distributions built in. |
Controls¶
The group box is headed Transformation.
| Label as shown | Control | Default | Valid entries | What it does |
|---|---|---|---|---|
| None | Radio button | Selected | n/a | The chart is built from the measurements as entered. |
| Use Johnson Transformation | Radio button | Not selected | n/a | Fits a Johnson family transformation. |
| Use Box-Cox Transformation | Radio button | Not selected | n/a | Fits a Box-Cox power transformation. All values used must be positive. |
| Use Optimal Lambda | Radio button | Selected as soon as Box-Cox is chosen | n/a | The power is fitted from your data. |
| Enter Custom Value | Radio button | Not selected | n/a | You supply the power instead of fitting it. |
| (lambda box) with the hint (-5 to 5) beside it | Text box, no label of its own | Empty | A number | The power to use. |
Hidden rather than disabled: the Use Optimal Lambda and Enter Custom Value buttons appear only while Use Box-Cox Transformation is selected, and the lambda box and its (-5 to 5) hint appear only while Enter Custom Value is selected. The Johnson transformation takes no parameters, so it reveals nothing.
Validation. The lambda box is checked against the range in the hint beside it. A value outside -5 to 5, or text that is not a number, is refused with a message box and the box goes back to what it held before:
The custom lambda must be a number from -5 to 5.
The range is the same one the fitted search uses, so a typed power outside it is a power the optimal fit would never have chosen. Earlier versions passed the value on as it stood and showed no message of any kind. The two shared math pages define the transformations and the range: Box-Cox Transformation and Johnson Transformation.
What a transformation affects¶
A transformation is not a display option. It changes the numbers the whole analysis is built from, so read the chart in transformed units afterwards.
| Affected | Not affected |
|---|---|
| The plotted points on every panel | Which points appear, and in what order |
| The center line, the control limits, and the zone lines | The rules that are switched on, and their thresholds |
| Every sigma estimate and its degrees of freedom | The sigma multiplier |
| The vertical axis, which is now in transformed units | |
| The values reported in the summary table | |
| Any specification limits, target, and the capability results computed from them |
The transformation is fitted once for the whole chart, and it is fitted on the baseline points only, not on all the data. With splits in effect, the single fit uses the union of the per-phase baseline windows, that is exactly the points that feed each phase's own center line and within-phase sigma, and the fitted transformation is then applied to every row of the chart. Individual phases never get their own transformation. Full detail: Splits and Phases.
Blank rows are handled slightly differently by the two transformations while the fit is being made: Box-Cox keeps a blank as a hole, so the moving range chain breaks in exactly the places it would have broken anyway, while the Johnson transformation drops blanks from the values it fits on.
With a transformation active on an EWMA chart, the smoothing runs on the transformed values: each plotted point is the exponentially weighted average of the transformed measurements, not the transformation of a smoothed measurement. The vertical axis and every value in the summary table are in transformed units.
Conditional behavior¶
| Condition | Effect |
|---|---|
| Any transformation is active | Manual control limits become unavailable. The Manual option is greyed out, and if it was already selected it reverts to the calculated limits. Limits typed by hand would be in the original units and there is nothing to convert them with. |
| Any transformation is active | The Historical Values boxes are greyed out, and any value already typed there is dropped for this run rather than converted. The typed text stays in the box and takes effect again when the transformation is switched back to None. |
| A transformation is active and a capability analysis is active | Split Control Limits is greyed out entirely and the split mode reverts to None. A capability analysis counts as active as soon as a lower or upper specification limit is entered. This combination can only arise on the charts that offer both a transformation and a Capability tab. |
| A transformation is active and splits are in effect | One transformation for the whole chart, fitted on the union of the per-phase baseline windows, applied to every row. Both the plotted values and the summary table report transformed results. |
| Estimate Limits From is set to a first-N count | The transformation is fitted from that baseline subset, not from all the data. Changing the count changes the transformation, and therefore changes every plotted value. |
| Box-Cox is selected | The Box-Cox transformation is defined for positive values only. Zero and negative measurements cannot be transformed this way. |
Two rows of that table cannot arise on EWMA. There are no manual control limits to make unavailable, and there is no Capability tab, so a transformation can never be combined with a capability analysis here. The row that does matter is the second one: turning a transformation on greys out the Historical Values group and drops what you typed there for that run.
Split Control Limits¶
A split divides the chart into consecutive sections, called phases, and gives each phase its own center line and its own control limits. Use it when the process changed on a known date: a new supplier, a tooling change, a settings change. The limits then step at the change instead of averaging across it.
Every phase estimates its own center line, its own sigma, its own control limits, and its own zone lines where the chart has them, and the out-of-control run counters restart at every phase boundary. Splitting never hides or moves a point: all points stay on the chart in their original order, and only the limits they are judged against change. For the formulas, and for the full list of what is shared across the chart versus estimated per phase, see Splits and Phases.
On EWMA a phase gets something extra of its own: the smoothing restarts at each phase boundary. The first point of a new phase starts again from that phase's own center line, and the band narrows and widens again from there, so a split chart shows one funnel per phase.
Controls, variable chart dialogs¶
The group box is headed Split Control Limits and holds three mutually exclusive mode buttons plus the editors each mode reveals.
| Label as shown | Control | Default | Valid entries | What it does |
|---|---|---|---|---|
| None | Radio button | Selected | n/a | One center line and one set of limits for the whole chart. |
| Split by column | Radio button | Not selected | n/a | A new phase starts at every row where the chosen column's value changes. |
| (split column list) | Drop-down list, no label of its own | Nothing selected | Any one column offered in the list | Names the column whose value changes drive the phases. |
| Manual break points | Radio button | Not selected | n/a | Phases start at the point numbers you type. |
| (break entry box) | Text box, no label of its own. Tool tip: "Enter a 1-based chart point (subgroup) number" | Empty | A whole number from 1 up to the number of plotted points | Holds one break number until you press Add. It is a staging box; typing in it alone changes nothing. |
| Add | Button | n/a | n/a | Moves the typed number into the break list, in ascending order, and inserts a matching label slot. |
| Remove | Button | n/a | n/a | Deletes the break selected in the list and its label slot. |
| (break list) | List box | Empty | n/a | Shows the break point numbers currently in effect. |
| Split Labels (shown beside each divider): | Label above one text box per phase | One empty box, named "Split 1" | Any text | The text drawn beside each divider on the chart. |
Hidden rather than disabled: the split column list appears only while Split by column is selected, and the break entry box, Add, Remove, the break list, and the whole Split Labels block appear only while Manual break points is selected.
Disabled: the entire group box is greyed out, and the mode reverts to None, when a transformation and a capability analysis are both active. See the conditional behavior table below.
Validation. Add refuses an entry it cannot use with a message box, and leaves what you typed in the box so you can correct it. Each way the entry can be wrong has its own sentence:
| What you typed | The message |
|---|---|
| Anything that is not a whole number | The split row must be a whole number. |
| A number below 1 | The split row must be a whole number of 1 or more. |
| A number at or beyond the number of plotted points | The split row must be less than the number of plotted points (120). |
| A break already in the list | The split row must be a row that is not already in the list. |
A split starts a new phase, so it has to fall strictly inside the data: row 1 is the first row that can start a second phase, and the last plotted point cannot start one at all. When the point count is not yet known to the dialog the upper check is skipped, so a break past the end of the data can be added and is dealt with later by the calculation.
Remove with nothing selected in the list does nothing and says nothing. There is still no check on how short a phase the break would create.
Earlier versions refused all four cases in silence, leaving the entry in the box with nothing on screen to say why the list had not grown.
What a break number means¶
A break number is the 1-based number of the plotted point that starts the new phase, the same number you read along the horizontal axis, not a worksheet row address. On charts plotted from subgroups it is the subgroup number, not the raw data row. On charts where one row is one point the two are the same.
Breaks must be different from each other and are kept in ascending order for you. A break at point 1 does nothing, because the first phase already starts there.
Split Labels¶
The Split Labels (shown beside each divider): block gives one text box per phase, so the number of boxes is always the number of breaks plus one. Each box is named Split 1, Split 2, and so on, in chart order. Those names are slot names, not the labels: they are fixed and cannot be edited. Whatever you type in the box beside a slot is what gets drawn on the chart. Leaving a box empty draws no label for that phase.
Adding a break inserts a new empty box at the position that changed rather than adding one at the end, so labels already typed stay with the phases they were written for.
The block appears only in the manual break mode, so there is no way to type phase labels when the phases come from a column. The labels are never read from the split column's values. Labels typed earlier in manual mode are kept when you switch to column mode, and they are still drawn beside the dividers; without them the dividers carry no labels.
The split column is claimed from the Data tab¶
The split column is chosen here, but it is one of the worksheet columns listed on the Data tab, and a column can hold only one job at a time. Choosing a column as the split column claims it: on the Data tab that column is greyed out and can no longer be checked as a measurement column, an X-axis label column, a subgroup ID, or a part column. Releasing it here gives it back. A column already claimed for one of those other jobs appears greyed in the split column list and cannot be picked. Neither the greyed check box on the Data tab nor the greyed list entry here explains which job is holding the column.
Selecting the column mode with no column chosen is not an error. It is a deliberate decision in the product, and it means do not split. Nothing is flagged, no message appears, and the chart is built with one set of limits, exactly as if the mode were None. The same is true of the manual break mode with an empty break list.
Minimum phase length, and two opposite outcomes¶
Every phase has to be long enough to estimate limits from. The minimum differs by chart family, and so does what happens when a break breaks the rule.
| Charts | Shortest phase allowed | A break that would create a shorter phase |
|---|---|---|
| Variable charts | 5 points | The whole analysis is rejected and no chart is produced. |
| Attribute and rare-event charts | 2 points | The break is dropped and the chart is built without it. |
That is the same user action with opposite outcomes: on an attribute chart a too-short phase costs you a divider, on a variable chart it costs you the chart.
On a variable chart the rejection reads The baseline must use at least 5 subgroups., the same message a short baseline gives, even though the cause is a phase that is too short.
The variable chart dialogs validate neither number. They do not check the minimum, and they do not check the arithmetic of the phase lengths your breaks imply, so a set of breaks can be accepted without complaint at entry and then fail when the chart is built. The attribute and rare-event dialogs do check the two-point rule as each break is added, with the message shown above.
A variable chart phase also fails the analysis when it holds fewer than five values that are actually present, even when the phase itself is long enough. Blank rows inside a phase therefore matter as much as its length.
On EWMA the five-point minimum is five plotted points, which is five subgroups when the data is subgrouped and five rows when the subgroup size is 1. The dialog does not check it, so a break that leaves a phase of four is accepted here and then stops the analysis with a message instead of producing a chart.
Conditional behavior¶
| Condition | Effect |
|---|---|
| Any split mode other than None is selected, on a variable chart | Manual control limits become unavailable. The Manual option is greyed out, and if it was already selected it reverts to the calculated limits. |
| A split mode is selected but produces no phases (column mode with no column, or manual mode with an empty list), on a variable chart | Manual control limits are still unavailable. The check looks at which mode button is selected, not at whether any phase boundary results. Setting the mode back to None restores the Manual option. |
| Any split mode selected, on an attribute or rare-event chart | No such restriction. Manual limits stay available, and the band you enter is drawn in every phase. |
| A transformation and a capability analysis are both active | The whole Split Control Limits group box is greyed out and the mode reverts to None. A capability analysis counts as active as soon as a lower or upper specification limit is entered. This can only arise on the charts that offer both a transformation and a Capability tab. |
| A transformation is active, with or without splits | One transformation is fitted for the whole chart and then applied to every row. Phases never get their own transformation. |
| Estimate Limits From, or Baseline Estimation, is set to a first-N count | There is one count for the whole chart. It is reused inside every phase and capped at that phase's length, so a short phase uses all of its own points while a longer phase in the same chart uses the count you typed. There is no per-phase count, and the count never reaches across a phase boundary. See Baseline and Limit Estimation. |
| Out-of-control rules are on | Every run and every window restarts at each phase boundary. A run in progress when a boundary arrives is abandoned, so a run of eight points that straddles a boundary cannot combine into one signal. The point-by-point test against the limits is unaffected, because each point is compared with its own phase's limits. |
| A historical mean or sigma is supplied | The one value you supply is applied to every phase. It is not a per-phase entry. |
Splits are edited after the fact in the Control Chart task pane as well, but only the manual break points and the labels. The column mode cannot be selected or cleared there.
No phases are produced under Subgroup ID column
When the Data tab takes its subgroups from a Subgroup ID column:, this group box produces no phase boundaries at all, in either split mode. The controls stay available and accept a split column or a list of break points, nothing is greyed out, and no message appears, but the finished chart carries one center line and one set of limits across the whole of it.
To split a chart into phases, set the subgroups with Constant size: or Subgroups across columns instead. Both of those honor either split mode normally.
The first two rows of that table are about manual control limits, which EWMA does not offer, so on this chart choosing a split mode takes nothing away.
Rules & Display Tab¶
The rule set on EWMA and CUSUM¶
These two charts offer Rule 1 only. The group box holds a single row, Rule 1 - Beyond K sigma from center, and it is ticked by default. No other rule has a row, and none can be turned on.
The reason is what these charts plot. Each point already carries the influence of the points before it, so the run and trend patterns the other rules look for are no longer independent evidence of anything.
EWMA draws one panel, so there is no question of which panel a rule runs on: the single test runs on the EWMA panel, against the widening limits drawn at each point.
Clearing the Rule 1 check box does not turn the test off
On this chart the check box is not read. The engine builds its own rule configuration for EWMA, with Rule 1 on, and it does that whether the box is ticked or cleared. Clear the box, press Finish, and points beyond the limits are still flagged and still drawn in the out-of-control color, while the dialog you just used says the test is off. Nothing reports the disagreement.
The one setting that does stop points being flagged is Control Limits set to None, which removes the limits and every out-of-control result together. There is no way to keep the band and switch the test off.
The sigma multiplier is also a rule setting
Rule 1 tests each point against the drawn control limits, so editing the number of standard deviations in the Control Limits section changes the Rule 1 test at the same time as it changes the limits. That is why Rule 1 has no threshold box of its own. It also decides whether rules 5 through 8 run at all: they read the zone lines, and zone lines exist only at three standard deviations, so at any other width those four do not run whatever their check boxes say.
Conditional behavior¶
| Condition | Effect |
|---|---|
| Control limits set to None, on a variable chart | No rule fires. Every rule result is cleared along with the limits, so no point is marked even though the rules are still ticked. |
| Control limits set to None, on an attribute or rare-event chart | The center line is still estimated, so rules 2, 3, and 4, and Rule 9 on a G chart, still fire and still mark points. Rule 1 cannot fire, because there are no limits to exceed. |
| Control limits set to Manual | The rules run against the band you entered, and the zone lines are derived from that band, still only on a location panel and still only at a sigma multiplier of 3. |
| Manual limits entered with the center left blank, on an attribute chart | There is no center line, so rules 2, 3, and 4 cannot fire. |
| Splits are in effect | Every run and every window restarts at each phase boundary. A run in progress when a boundary arrives is abandoned and cannot combine with the run that starts after it. |
| A point is blank or excluded | Runs break at the gap rather than bridging it. For the two window rules, rules 5 and 6, a blank invalidates every window that overlaps it, not just the run. |
| A rule is unticked | Its threshold value is kept and its box is locked. Nothing is lost by turning a rule off and on again. |
| The sigma multiplier is moved off 3 | No zone lines are drawn on any panel, and rules 5 through 8 do not run. The dialog greys its Show zone lines box and says so, and the Rules in effect block on the worksheet lists only the rules that ran. Z-MR is fixed at 3 and is never affected. |
| A capability analysis or a transformation is active | Neither affects the rules. The rule set, the thresholds, and the panels checked are the same. |
See also: Out-of-Control tests.
Three rows of that table do not apply to EWMA: the two manual-limit rows, because there are no manual limits, and the row about unticking a rule, because of the behavior described in the warning above. Read the rows about None, about splits, and about blank points, which do apply.
A point that broke the test is drawn in a different color from the rest, and that is the whole of the report. Nothing in the chart or the worksheet labels the point.
Display Options¶
The Display Options group box holds two features: the sigma zone lines, and reference lines of your own at Y values you choose. The variable charts take it from a shared control. The p, np, Laney P', c, u, Laney U', G, and T dialogs repeat the same block for the reference lines, but carry no zone line checkbox, because those charts draw no zone lines.
This chart draws no zone lines, so the dialog shows no zone line checkbox at all, and the Analysis Pane hides its own Show zone lines checkbox when the chart is selected.
The reason is on the Math Details page: an EWMA band widens point by point, so a fixed fraction of it is not a fixed number of standard deviations, and the only test this chart offers reads the limits rather than the zones.
Reference lines¶
A reference line is a horizontal line at a Y value you supply: a specification limit, a target, a contract threshold, anything you want the points compared against. Until you add one the editor reads "No reference lines configured. Use the row editor below to add Y-value reference lines."
Each reference line is one row with four editable cells and a remove button:
| Cell | What it sets | Default |
|---|---|---|
| Value | the Y position of the line | 0 |
| Label | the legend entry for the line | blank |
| Color | the line color, picked from a color button | red |
| Thickness | the line weight, chosen from 0.5, 1.0, 1.5, and 2.0 | 1.0 |
| X | removes that row. Its tooltip reads "Remove this reference line" | n/a |
Value is not validated. Type something that is not a number and the cell keeps the value it had. A blank Label keeps the line out of the chart legend; the line is still drawn, under a generated name built from its value. The editor scrolls on its own once the list of rows is taller than the space allowed for it.
- Add Reference Line: appends a blank row and refreshes the chart preview by itself. Removing a row with its X button does the same.
- Apply: commits edits to Value, Label, Color, and Thickness, which do not refresh the preview on their own. Its tooltip reads "Refresh the chart preview to reflect any pending reference-line edits". The button appears only once at least one reference line exists, and does nothing when there is nothing outstanding to commit.
So adding and removing lines is immediate, while editing an existing line needs Apply.
Reference lines are drawn on the location panel only
Reference lines appear on the panel that plots the measurement itself: the Xbar panel, the Individuals panel, the EWMA panel, or the Z panel of a Z-MR chart. The dispersion panels (Range, Standard Deviation, Moving Range) are in different units, so they never carry reference lines. The location panel's Y axis is stretched to include every reference line value, so a line outside the data still shows.
This has a consequence on the single-panel dispersion charts. R only, S only, and Moving Range only draw no location panel, so a reference line entered in the editor never appears anywhere on the chart. The editor still accepts it and still lists it. If you need a reference line against a range or a standard deviation, there is no dialog option that will draw one.
CUSUM is the exception: it is a single overlay chart rather than stacked panels, and it draws reference lines on that chart.
EWMA has one panel and it is the location panel, so every reference line you enter is drawn, and the vertical axis is stretched to include it.
Layout you do not choose¶
Nothing else about the appearance of the chart is a dialog option. These are fixed:
- Which panels are drawn is not part of Display Options. On Xbar-R, Xbar-S and I-MR it is the Chart: radio group that sits above the tabs, and the ribbon entry you started from only decides which of those buttons is already selected when the dialog opens. You can change it there without closing the dialog. Every other control chart draws a fixed set of panels and offers no control over it.
- Panel titles are composed for you from the panel type, the group, and the measurement column name, with empty parts dropped.
- Plot order is fixed, and it is also the front-to-back order: the data line, then the zone lines on a panel that has them, then the upper and lower control limits, then the center line, then the reference lines. The zone lines are added before the limits deliberately, so they stay in the background.
- Axes are set for you. The X axis is numbered by plotted point and titled Subgroup. The Y axis is scaled to fit every plotted value, including the zone lines on a panel that draws them and, on the location panel, the reference lines; dispersion panels are held at or above zero. When a chart has more than one panel the panels are aligned with each other so their points line up vertically.
Once the chart exists, zone lines that were drawn can be shown or hidden from the Analysis Pane without running anything again. Zone lines that were never drawn cannot be brought back from the pane: the pane hides its checkbox, and the only way to get them is to modify the analysis, set the control limits to 3 standard deviations, and run it again. Reference lines are not in the pane at all, so to add, edit, or remove one you modify the analysis and run it again.
Task Pane¶
The Control Chart pane that opens beside a finished chart has its own page: Task Pane.
When a number box refuses what you typed¶
A refused number box now tells you why. Type something this box cannot use, then leave it, and a message box opens naming the box and what it wanted:
The box name must be what it wanted.
The title is the analysis name, for example "G Chart" or "Type 1 Gage Study". A box that wants a proportion adds a second sentence, "It is a proportion, not a percent, so type 0.05 for 5 percent.", because typing 5 for five percent is the mistake it is there to catch. The box then goes back to the value it held before, so nothing is lost and nothing wrong is carried into the run.
Text that is not a number at all gets the same message. Type a word, a stray letter beside a figure, or nothing but a minus sign into any number box on this dialog, leave the box, and a message box opens reading:
The box name must be a number.
The box then shows the value it held before. Before this, a box given text simply would not take it: the entry was dropped on the way to the setting, so none of the rules below could be reached and nothing was said. The box looked as though it had accepted what was typed until the moment the focus moved.
The dialog no longer runs the analysis while a setting is unusable. Changing almost any option on this dialog redraws the chart straight away, without waiting for Finish. That redraw now reads the same rule the Finish button reads. When something on the dialog is wrong, the redraw does not happen and the dialog stays open, with the reason printed beside the Finish button.
Where the problem is two settings that contradict each other, and you have finished making both choices, a message box says so as well. Where it is a box you have not finished filling in, the dialog waits quietly instead, with the reason still beside Finish.
Nothing you typed or clicked is undone. The setting you just changed stays changed, the number you typed stays in its box, and the same sentence stays on screen beside the Finish button, which is switched off until you resolve it. Change the other half of the pair, or put the number back, and the chart redraws as usual.
A message box is shown once per reason, not once per redraw. A single click can raise a dozen redraws internally, and repeating the same message box a dozen times would be worse than saying it once, so a second attempt that hits the same reason shows no message box. That is not the dialog ignoring you: the reason is still printed beside Finish the whole time.
This matters because these two-setting rules were previously unreachable. The sentence beside Finish was written and Finish was switched off, but the redraw ran the analysis anyway without ever consulting either, so the chart was drawn from the settings the dialog had already refused.
These boxes are shared by every variable control chart dialog, so they refuse in the same words wherever they appear. A dialog that does not offer a control does not show its box at all.
| Box | What it must be |
|---|---|
| Number of standard deviations (k): | a number greater than 0 |
| Mean: and Mean (between): under Historical Values | a number, or blank. A process mean may be negative, so nothing is ruled out except text. |
| Standard deviation (overall):, (within): and (between): under Historical Values | a number greater than 0, or blank |
| Manual UCL, Center and LCL, per panel | a number, or blank, and in increasing order down the row. The message names the panel: "The manual UCL for the Individuals panel must be greater than the manual center." |
| First N subgroups: count | a whole number of 5 or more |
| Moving-range length (L): | a whole number from 2 to 100 |
| The custom Box-Cox lambda box, hinted (-5 to 5) | a number from -5 to 5 |
n = beside Constant size: |
a whole number from 2 to 100, or from 1 to 100 on a chart that accepts individual values |
| N = on the rules that count consecutive points (2, 3 and 4) | a whole number of 2 or more. The message names the rule: "The rule 3 N must be a whole number of 2 or more." |
| N = and M = on the N-of-M rules (5 and 6) | N a whole number of 1 or more, M a whole number of 2 or more, and N no greater than the M beside it |
| N = on the Nelson rules 7 and 8 | a whole number of 1 or more |
| Sigma = on a Westgard rule | a number greater than 0 |
| LSL:, USL: and Target: on the Capability tab | a number, or blank, with LSL below both the others and USL above both. A box is compared only against a box that holds a value. |
| Sigma tolerance under the Custom capability standard | a number greater than 0 |
| Confidence level | greater than 0 and less than 1, and the message adds "It is a proportion, not a percent, so type 0.95 for 95 percent." |
| The histogram anchor value | a number, or blank |
| The histogram bin count | a whole number from 1 to 250 |
| The histogram bin width | a number of 0.001 or more, or blank |
| The split row box beside Add | a whole number of 1 or more, below the number of plotted points, and not already in the list |
EWMA has no manual limits grid, because its limits change at every point. It accepts a constant subgroup size of 1, so its n = box takes a whole number from 1 to 100. One box is its own:
| Box | What it must be |
|---|---|
| EWMA weight (λ): | a number greater than 0 and at most 1. The value 1 is allowed and plots the values themselves; 0 is not, because a weight of zero would never move the line off its starting point. |