Académique Documents
Professionnel Documents
Culture Documents
to
LMSchartmaker
1997-2011 Medical Research Council, UK
Table of Contents
Foreword 0
Part I Introduction
1 What is LMSchartmaker ................................................................................................................................... 4 2 LMSchartmaker Light ................................................................................................................................... 4 3 LMSchartmaker Pro ................................................................................................................................... 4 4 Notes ................................................................................................................................... 4 5 Contact Us ................................................................................................................................... 4
1 Data Display ................................................................................................................................... 7 2 Data Summary ................................................................................................................................... 7 3 Data Editing ................................................................................................................................... 9 4 Data Selection ................................................................................................................................... 9
Part IV Modelling
11
1 Model Selection ................................................................................................................................... 11 2 Model Fitting ................................................................................................................................... 11 3 Automated Model Fitting with the Loop ................................................................................................................................... 12 4 Model Monitoring ................................................................................................................................... 12 5 Model Checking ................................................................................................................................... 13 6 Choosing edf ................................................................................................................................... 13 7 Choosing the Age Option ................................................................................................................................... 14 8 Choosing the Deviance Measure ................................................................................................................................... 14
14
1 The System File ................................................................................................................................... 14 2 How to Save ................................................................................................................................... 15 3 Save the ................................................................................................................................... 15 Model 4 Save the ................................................................................................................................... 15 Model or Z Scores 5 Export the L,M,S and/or Centiles ................................................................................................................................... 15
16
Contents
II
Data Plot ......................................................................................................................................................... 17 Z Scores ......................................................................................................................................................... 17 Detrended Q-Q Plot ......................................................................................................................................................... 18 Q Tests for Fit ......................................................................................................................................................... 19 L Curve ......................................................................................................................................................... 19 M Curve ......................................................................................................................................................... 20 S Curve ......................................................................................................................................................... 21 Centiles ......................................................................................................................................................... 21 M Curve Velocity ......................................................................................................................................................... 22 M Curve Acceleration ......................................................................................................................................................... 23
Part VII Print Active Window Part VIII Example of Free Format with LMSchartmaker Light
24
24
1 Example ................................................................................................................................... 24 of Free Format 2 Fit Model................................................................................................................................... 24 Step by Step 3 Open Data in Free Format ................................................................................................................................... 25 4 Display Data ................................................................................................................................... 26 5 Rename Measure to Weight ................................................................................................................................... 26 6 Model of ................................................................................................................................... 27 Weight (030503o) 7 Q Tests for Fit Weight (030503o) ................................................................................................................................... 27 8 Model of ................................................................................................................................... 28 Weight (030603o) 9 Q Tests for Fit Weight (030603o) ................................................................................................................................... 28 10 Model of ................................................................................................................................... 29 Weight (030603r) 11 Q Tests for Fit Weight (030603r) ................................................................................................................................... 29 12 Model of ................................................................................................................................... 30 Weight (030703r) 13 Q Tests for Fit Weight (030703r) ................................................................................................................................... 30
31
1 Example ................................................................................................................................... 31 of Fixed Width 2 Fit Model................................................................................................................................... 31 Step by Step 3 Open Data for Fixed Width ................................................................................................................................... 31 4 Rename Measure to BMI ................................................................................................................................... 32 5 Display Data ................................................................................................................................... 33 6 Model of ................................................................................................................................... 33 BMI (030503o) 7 Model of ................................................................................................................................... 34 BMI (030503t, 0.5,0) 8 Model of ................................................................................................................................... 36 BMI (030803t, 0.5,0)
III
Part X Preferences
37
1 Summary ................................................................................................................................... 37 2 Centiles ................................................................................................................................... 37 3 Limits ................................................................................................................................... 38 4 Weighting ................................................................................................................................... 39 5 Register ................................................................................................................................... 39
40
1 Missing Values ................................................................................................................................... 40 2 Power and Offset ................................................................................................................................... 40 3 Default Name ................................................................................................................................... 40 4 edf ................................................................................................................................... 40
41
1 Deviance................................................................................................................................... 41 2 Z scores ................................................................................................................................... 41 3 Q Tests ................................................................................................................................... 41 4 Chi Statistic for Q Tests ................................................................................................................................... 41 5 Q Statistic for Q Tests ................................................................................................................................... 42 6 Detrended Q-Q Plot ................................................................................................................................... 43
Part XIII Acknowledgement Part XIV Statistical References Part XV Window Stacked and Customised Index
43 43 44 0
1 Acknowledgement ................................................................................................................................... 43
1 References ................................................................................................................................... 43
Introduction
1
1.1
Introduction
What is LMSchartmaker
LMSchartmaker is a program to fit smooth centile curves to reference data using the LMS method. Reference centile curves show the distribution of a measurement as it changes according to some covariate, e.g. age. The LMS method summarises the changing distribution by three curves representing the median (M), coefficient of variation (S) and skewness (L), the latter expressed as a Box-Cox power. Using penalised likelihood the three curves can be fitted as cubic splines by non-linear regression, and the extent of smoothing required can be expressed in terms of smoothing parameters or equivalent degrees of freedom. The process involves five stages: data entry, model fitting, graphical display, model checking and model saving, with the last four stages repeated as often as necessary. Data editing and graph editing are also available from menu Edit. In addition LMSchartmaker Pro allows you to use weighting, selecting group and goodness of fit tests. Start LMSchartmaker by clicking Start, Programs and LMS. To exit click Exit from the File menu, that is click File|Exit.
1.2
LMSchartmaker Light
LMSchartmaker Light is the freeware version of LMSchartmaker. It contains data entry, model fitting, graphical display and model checking. You can download LMSchartmaker Light from the web. There is also an extended version of the LMSchartmaker, LMSchartmaker Pro, for which a licence key is needed. Note LMSchartmaker Pro can open a system file saved from LMSchartmaker Light, but LMSchartmaker Light cannot open a system file saved by LMSchartmaker Pro.
1.3
LMSchartmaker Pro
LMSchartmaker Pro is the extended version of LMSchartmaker. It has all the facilities of LMSchartmaker Light, but in addition LMSchartmaker Pro fits models taking into account sample weights and group selection. It also provides improved goodness of fit tests including Q Tests. Contact Health for All Children to obtain a key code to convert LMSchartmaker Light to Pro. Then click on Edit|Preferences|Register to register the key code and upgrade the program.
1.4
Notes
Number of decimal places for age is controlled by LMSchartmaker to avoid numerical instability and speed up the calculations. Non-postive values for measurement, weighting or selector are treated as missing. Any missing values for age or Xaxes should not be coded as negative or zero because LMSchartmaker allows age to be zero or negative. Maximum sample size is 100000 by default. For special need of large sample size please contact us.
1.5
Contact Us
You can contact us in several ways: contact Tim Cole tim.cole@ich.ucl.ac.uk for ordering Health for All Children for technical support (LMSchartmaker Pro users only) lms.help@hotmail.co.uk Note We suggest you use the most recent version of your browser, and periodically check for new versions of LMSchartmaker.
2
2.1
Data Entry
Opening an Existing File
File|Open allows you to open an existing ext file (e.g. *.txt) or LMSchartmaker system file (*.lms) Click on File|Open to bring up an Open dialog box. Open a text file - locate the subdirectory in the Look in box, select Text file (*.txt) or All files (*.*) in the Files of type box, and double-click the file you want to open. This brings up a Data Type dialog box , where you specify the location of the age or time and measure for both LMSchartmaker Light and LMSchartmaker Pro. In LMSchartmaker Pro you can
Data Entry
2.2
3
3.1
3.2
Data Summary
Original Data - shows the information for original data. They are data file name, location of age and measures, total number of records and the number of missing values. The location contains the field number of variables for data in free format and position for both start and end of variables for data in fixed width. LMSchartmaker Pro shows the location for weighting and selector if they are available. LMSchartmaker Pro will also show the name and number of groups if selector is considered. See below for an example. LMS Data - shows the information for data available for LMS model fitting. They are the number of positive points,
distinct ages, range of age and measure. When selector is available LMSchartmaker Pro shows the name and number of groups. See below for an example.
3.3
Data Editing
LMSchartmaker allows you to view and edit data. There are two ways to edit data: Editing individual data values - click in the cell containing the value, type the new value and press <return> or click elsewhere in the window. Editing highlighted data - File|Cut and File|Copy allow you to cut or copy the highlighted data from Data window to the clipboard while File|Clear removes the highlighted data. File|Paste allows you to paste data from the clipboard into the data sheet where the cursor is located. The data set will normally have been copied to the clipboard from another application such as a database, spreadsheet or word processor. To copy data from such an application highlight the data required and use the application's Edit|Copy facility. Edit|Select All highlights all the data. Note Editing data will re-start model fitting. Negative or zero in Weighting is treated as a missing value. Blank in Selecting is considered as missing if you select groups.
3.4
Data Selection
LMSchartmaker Pro allows you to select part of your data by including a column with group codes in the Data Window. Group can be coded either as a number or character with up to 8 digits. For example you can code 1 for male and 2 for female or m for male and f for female. A blank in the group is treated as a missing value. By default LMSchartmaker Pro uses all the data. In addition both LMSchartmaker Pro and Light allow you to select data in the range of age or measurement. To select data click the Select button in the Data Window, to bring up the Select Data dialog with two tabs, Range and Group. Group tab is only enabled for LMSchartmaker Pro with group codes. Range - specify the range in the corresponding boxes and click OK. Group - Highlight the group(s) you want in the Group list box and click OK.
10
Note The Reset button lets you re-set the previous selection and the Select All button brings you back to use all the data.
11
4
4.1
Modelling
Model Selection
The model is specified using the Model menu. This involves selecting the right age scale and choosing edf values for L, M and S. There are three choices for the age scale depending on the shape of the M curve: Original Age is suitable for simple M curves - this is the default. Rescaled Age is appropriate for more complicated M curves like weight or height throughout childhood. It works best where the M curve is monotonic, i.e. increasing (or decreasing) throughout the age range. Transformed Age is best for complex M curves that are non-monotonic, i.e. which rise and fall at different times, such as body mass index or skinfold thickness. There are three aspects to choosing proper edf values: Choose the edf in the order: M first, then S and then L. After fitting the initial model fine-tune the edf for the final model. Choose the edf so that: M edf > S edf > L edf. Larger samples need larger edf. If in doubt use a smaller not a larger edf. A choice of fitting one model at a time or a series of models: The check box Loop allows you to switch on or off automated model fitting. See Model Fitting and Model Checking to help you select a model.
4.2
Model Fitting
Click the Model menu to show the Model dialog box. Select the age scale and edf for L, M ans S in the model, then click Fit. The model is fitted and the results displayed: The deviance and edf are in the first line of the lower part of the Model dialog box. The second and third lines contain values for previous model (if fitted) and the difference between present and previous. Four graphs are displayed: L Curve, M Curve, S Curve and Centiles. Graphs for Z scores, Detrended Q-Q Plot and Q Tests (the latter two for LMSchartmaker Pro only) can be obtained from Graph menu. The model is also summarised in the Monitor dialog box, obtained by clicking the Monitor button. To sepcify link for S or M click Edit|Preferences|Link. The default is Identity Link for both S and M. Click on the deviance label in the Model dialog to cycle through various measures of deviance. The aim is to find a model that minimises the measure. Note If the model fails to fit a warning is given and no deviance is presented.
Modelling
12
4.3
4.4
Model Monitoring
For each fitted model LMSchartmaker keeps a record of the deviance, edf for L, M and S, the number of cycles, the scale of age, power and offset. To view the records click on the Monitor button in the Model dialog box to bring up the Monitor window. It contains up to 500 models: The first column contains the deviance measure (which appears in the lower part of the Model window, and which is changes by clicking on its label there). To sort the models in ascending order of the deviance measure click Yes at the foot of the Monitor window. Sorted this way the first model is the "best". To rank the models in their order of fitting click No. To plot the deviance measure against edf (only for models fitted with Loop) click Plot. Three curves are plotted in the same window, showing the deviance measure versus edf for L, M and S separately for the current Loop. To re-fit a model select its row and click Re-fit. To print the records click Print. To clear the records click Clear. To copy the records click Edit|Copy. To copy records click Edit|Copy when the Monitor dialog is active. To copy records click Edit|Copy when the Monitor dialog is active.
13
4.5
Model Checking
The Model window indicates the fit of the current model as measured by the deviance. When the model changes, the Model window shows the change in edf and the corresponding change in deviance. A cumulative log of the models fitted so far, their deviance and parameters, can be found in the Monitor dialog box. The following are useful: The Z Scores Graph is useful to check outliers. The comparison of current and previous fitted L, M, S curves and centiles. See graphical display for details. The Detrended Q-Q Plot is a detrended form of the Q-Q plot for use as a global diagnostic. The Q Tests for Fit help test goodness of fit when judging a model. A model may be considered adequate if the Q Statistic curves for L, M and S are within the range -2 and +2. Click here to display these graphs for model checking.
4.6
Choosing edf
The order for choosing edf values Choose edf values for L, M and S in the following order: Optimise the M curve by increasing and/or decreasing the edf by 1 until the change in deviance is small. For small datasets (e.g. n < 500) the default edf of 5 may be adequate. For larger datasets (n > 10,000) the edf may reach 15 or more. Optimise the S curve. In many cases the default of 3 edf will be sufficient, though edf up to 10 or more may be needed for larger datasets. Optimise the L curve. The default of 3 edf may be too large in many cases. Try a value of 0, i.e. no skewness adjustment, then 1, which is a constant adjustment at all ages. Then increase edf in steps of 1. Values exceeding 6 are unlikely except for large datasets covering a wide age range. The general rule for choosing edf values The simplest rule for choosing the edf is: use the Loop and minimise the selectd deviance measure. Alternatively, here are some guidelines: Change the edf 1 unit at a time. See how the curve changes shape as the edf change - the curve should be both smooth and a good fit. To see subtle changes in the curve double click in the curve graph window to compare current and previous curves. Watch the change in deviance in the Model window - a change of less than 4 per unit change in edf is modest, 10 or more is substantial. If in doubt, use smaller rather than larger edf.
Modelling
14
Sometimes, to clarify the underlying curve shape, it is worth using large edf, say 20+. The resulting curve will be very rough, but it will show whether alternative fits with fewer edf are biased or not. Larger samples need larger edf. In general, M edf > S edf > L edf. If you have LMSchartmaker Pro use the Detrended Q-Q Plot and Q Tests for Fit.
4.7
4.8
5
5.1
Model Saving
The System File
A system file is identified by the extension of LMS or lms, for example, TEST03050T.LMS or test030503t.lms. The LMSchartmaker gives a default name when saving a model to a file. The LMSchartmaker system file stores the parameters of the fitted model and information about the data used, which LMSchartmaker needs to retrieve the model when it opens the system file later. It contains three parts: 1. The first part contains two lines. The first line gives edf of the L, M and S curves. The second line contains: values of the age scale setting - age scale, power and offset. The value of age scale is coded as: 0 - original age, 1
15
5.2
How to Save
You can save a fitted model via the menu File: File|Save a system file (*.lms or *.LMS), from which LMSchartmaker can later retrieve the fitted model. For more information, click here. File|Save As either a system file (*.lms or *.LMS) or a text file (*.txt) with Z scores calculated by the fitted model. For details, click here. File|Export to save the fitted L, M, S curves and/or centiles with specified age intervals to a file (*.ept or *.EPT). Click here for more information. All the saved files can be read by NotePad, WordPad, Microsoft Word, Excel etc.
5.3
5.4
5.5
Model Saving
16
choice of saving With Head Information, which is only available when you opened a text file. It records the field for age, measure, weighting or selector in the original text file. The Head information is given at the beginning of the export file. choice of Set Default allows you to save the Export Option in the LMSchartmaker. Age Units for the setting of age units in export files includes Data and Export let you specify the data units, unit (default), year, month, week or day in the original or export file respectively. For example if the original age unit is years you can choose months in the Export box if you prefer. For every 1 unit of age the program exports 1 record by default, but you can change it. For the example above, with data in years exported as months, if you type 1 in the For every box and type 2 in the Export box, the records in the export file are for 0.5 month intervals.
6
6.1
6.1.1
17
6.1.2
Data Plot
A scatter plot of the data against age appears when the data are read in, e.g. using menu File|Open. If it is subsequently closed the same plot can be obtained by selecting Data on the Graph menu. There are two options to displaying the Data Plot: Show centile curves. Double click in the Data Plot window to show or hide centile curves for the current model (where available) superimposed on the data. Identify data points. First click in the Data Plot window to make it active. Then select Edit|Graph and click On in the Identify tab. This turns Identify status on. Now click on a point in the Data Plot window to display its sequence number next to the point, and the Data window scrolls to show the point. This is useful to examine or alter outliers. If the identify tab is on, the sequence number also appears in the Detrended Q-Q Plot or Z Scores graph if open. For details of editing a graph click here. See below for an example of Data Plot. Note 'Add centile curves' and 'Identify data points' do not operate together. If Identify status is on, double clicking does not show or hide centile curves.
6.1.3
Z Scores
Once a model has been fitted, the data can be converted to Z Scores. A graph of the Z scores plotted against age is obtained by selecting Z Scores on the Graph menu. There are two options to displaying the Z Score graph: Show centile curves. Double click in the Z scores graph window to show or hide centile curves for the current model (where available) superimposed on the individual Z scores. The centile curves appear as horizontal straight lines. Identify data points. First click in the window to make it active. Then select Edit|Graph and click On in the Identify tab. This turns Identify status on. Now click on a point in the Z scores graph to display its sequence number next to the point, and the Data window scrolls to show the point. This is useful to examine or alter outliers. If the identify tab is on, the sequence number also appears in the Data Plot graph or Detrended Q-Q Plot if open. For details of editing a graph click here. See below for an example of Z Scores Plot. Note 'Show centile curves' and 'Identify data points' do not operate together. If Identify status is on, double clicking does not show or hide centile curves.
18
6.1.4
19
6.1.5
6.1.6
L Curve
The L curve is the smoothed curve of the Box-Cox power transformation plotted against age. The L curve appears automatically when a model is first fitted. If it is subsequently closed the same plot can be obtained by selecting L Curve on the Graph menu, so long as a model is available. Double click in the graph window to show or hide the L curve for the previous model, where available. The previous curve appears in blue, with the current curve in black. This highlights the effect on the L curve of a change in the model. For details of editing a graph click here. See below for an example of L Curve.
20
6.1.7
M Curve
The M curve is the smoothed median of the measurement plotted against age. It corresponds to the median curve on a conventional growth chart. The M curve appears automatically when a model is first fitted. If it is subsequently closed the same plot can be obtained by selecting M Curve on the Graph menu, so long as a model is available. There is two M curve options: Show previous model. Double click in the graph window to show or hide the M curve for the previous model, where available. The previous curve appears in blue, with the current curve in black. This highlights the effect on the M curve of a change in the model. Alter axes. See (jump) Edit Graph for options to alter the M curve axes. For details of editing a graph click here. See below for an example of M Curve.
An example of M Curve
21
6.1.8
S Curve
The S curve is the smoothed curve of the measurement's coefficient of variation as it changes with age. The S curve appears automatically when a model is first fitted. If it is subsequently closed the same plot can be obtained by selecting S Curve on the Graph menu, so long as a model is available. Double click in the graph window to show or hide the S curve for the previous model, where available. The previous curve appears in blue, with the current curve in black. This highlights the effect on the S curve of a change in the model. For details of editing a graph click here. See below for an example of S Curve.
6.1.9
Centiles
Centile curves are smoothed curves calculated from the L curve, M curve and S curve, which in turn depend on the current model. The Centiles Graph appears automatically when a model is first fitted. If it is subsequently closed the same plot can be obtained by selecting Centiles on the Graph menu, so long as a model is available. There is two options for display the graph: Show previous model. Double click in the graph window to show or hide the centile curves for the previous model, where available. The previous curves appear in blue, with the current curves in black. This highlights the effect on the centiles of a change in the model. Select centiles. Edit|Preferences|Centiles changes the number or spacings of the centiles. For details of editing a graph click here. See below for an example of Centiles Graph.
22
23
6.2
6.2.1
24
6.2.2
Graph Resizing
A graph is sized with the window, so to resize the graph resize the graph window. Use the maximum button at the upperright corner of the graph window to maximise the graph.
6.2.3
Graph Copy
Graphs can be copied to other applications, for example Word or Excel. The graphs are those of Data, L Curve, M Curve, S Curve, Centiles, Z Scores, Q Tests for Fit, Detrended Q-Q, M Curve Velocity and M Curve Acceleration. You may need to edit the graph or resize a graph to obtain appropriate and clear labels and titles. Click the graph you want to copy and click Edit|Copy to copy the active graph window to the clipboard. Then you can paste it to other applications.
7
7.1
8
8.1
8.2
25
8.3
26
8.4
Display Data
Once data have entered LMSchartmaker displays Data window and Data Plot. The Goto, PgU, PgDn and Home buttons in the Data Window allows you to display the data at different location. Data summary is available by click Summary. Click here for details of data display. Click Rename to load the Rename dialog box.
8.5
27
8.6
8.7
28
8.8
8.9
29
8.10
8.11
30
8.12
8.13
31
9
9.1
9.2
9.3
32
Note LMSchartmaker Light provides the same summary except weighting and selector.
9.4
33
9.5
Display Data
Now the title Measure is changed to be BMI in both Data Window and Data Plot window.
Note LMSchartmaker Light provides the same summary except weighting and selector.
9.6
34
9.7
35
36
9.8
37
9.9
10
10.1
Preferences
Summary
Edit|Preferences allows you to set preferences for: Centiles to specify the number and SD spacing of centiles. For details click here. Limits to set the values of inner cycles, outer cycles, inner tolerance, outer tolerance and edf tolerance. For more information, click here. Weighting provides a choice of using sample weights. Click here for more information. Register to show register status or to upgrade LMSchartmaker Light to Pro. Click here for details.
10.2
Centiles
The centiles can be specified in two ways: either equally spaced on the Z score scale, or as a defined set of centiles. The default setting is 7 centiles equally spaced two-thirds of a Z score apart. The common set of 7 centiles, 3rd, 10th, 25th, 50th, 75th, 90th and 97th, is also preset. The centiles appear in the Centiles graph, and optionally in the Data and Z Scores graphs, and can be exported by File|Export. To change the centiles click on Edit|Preferences|Centiles: To specify equal spacing on the Z score scale click the button Centiles with equal SD spacing. Then enter the required number of centiles and their spacing. For example type 7 for the Number of centiles and 0.6667 for the Space between centiles to specify seven centiles spaced 0.6667 SD apart. See below for the Select Centiles dialog box. To specify unequally spaced centiles click the button Centiles with unequal SD spacing. Then list the required centiles in the cells under Centile. For example type 3, 10, 25, 50, 75, 90, 97 to obtain the centiles. To find out the Z scores corresponding to the centiles click in the Centile cell. See below for the Select Centiles dialog box. Note
Preferences
The defined centiles remain the default until you change them.
38
10.3
Limits
In addition the fitting process involves two nested iterations, and the user can adjust the parameters governing the iterative process. However there is little point changing them unless you are very sure you know what you are doing. Clicking on Edit|Preferences|Limits brings up the Limits dialog box, which allows you to change the following parameters for model fitting: Inner cycles the number of inner iterations is set to be 10 as default. Outer cycles the number of outer iterations is set to be 20 as default. Inner tolerance the criterion for inner convergence, 0.001 is default. Outer tolerance the criterion for outer convergence, 0.01 is default. edf tolerance the criterion for fitting edf, 0.05 is default.
39
10.4
Weighting
LMSchartmaker Pro can take account of sample weights, in model fitting. Usually all data points are of equal importance, but in complex survey designs data points can have different probabilities of being selected, indicated by their sample weight. A larger sample weight (not to be confused with body weight) indicates greater importance. A correct analysis needs to take the sample weights into account. If sample weights are not specified, weights of '1' are used for all the data. LMSchartmaker Light also uses equal weights throughout. Clicking on Edit|Preferences|Weighting brings the Weighting dialog box. There are three options: Scale sample weights to sum to sample size (default) Leave sample weights unscaled Ignore sample weights Note Zero or negative sample weights are ignored when fitting the model.
10.5
Register
Clicking Edit|Preference|Register shows register status in LMSchartmaker Pro and brings up the Register and Upgrade dialog box in LMSchartmaker Light: a Register Status dialog shows the registered user name and registered ID in LMSchartmaker Pro. a Register and Upgrade dialog allows you to buy a key code on internet and upgrade and register to LMSchartmaker Pro with name and key code. Note LMSchartmaker Pro can not be converted to LMSchartmaker Light.
Preferences
40
11
11.1
Glossary of terms
Missing Values
Negative values for measures, and weighting or selector if inclusive.
11.2
11.3
Default Name
By default, the name of a system file (*.LMS or *.lms) is given by the name of the data file followed by the values of edf. For example, test_030505o.lms, where each two digits of 030505 present the edf for L, M and S curves correspondingly and the character of o following denotes original age scale while the r and t for rescaled and transformed age option if you selected.
11.4
edf
edf (e.d.f.) stands for 'equivalent degrees of freedom'. The edf of each L, M and S curve is a measure of its complexity. 0 edf indicates a fixed constant whose value is specified in the box, e.g. L = 1 forces a normal distribution 1 edf relates to a constant 2 edf corresponds to a straight line 3 edf gives a simple curve like a quadratic 4 or more edf indicates progressively more complex curve shapes
41
12
12.1
Glossary of Statistics
Deviance
Five types of deviance measure are available for checking the goodness of fit: Penalized deviance, P.Deviance = -2 x penalized log likelihood Schwarz Bayesian Criterion, SBC = G.Deviance + edf_total x log(n) Generalised Akaike Information Criterion, GAIC(3) = G.Deviance + edf_total x 3 Akaike Information Criterion, AIC = G.Deviance + edf_total x 2 Global deviance, G.Deviance = -2 x global log likelihood where edf_total is the sum of the edf for L, M and S, and n is the sample size.
12.2
Z scores
The Z score of a child's measurement y is calculated from the L, M and S curves, using values appropriate for the child's age and sex. Two formulae are relevant depending on the value of L:
Z =
(y / M )L -1 , LS
if L 0
Z =
log( y / M ) , S
if L = 0
If the model is adequate the Z scores are approximately Normally distributed, with mean 0, standard deviation 1 and no skewness. If the value of L is 1, the original data are Normally distributed and the conventional Z score formula applies:
Z =
y-M M S
12.3
Q Tests
LMSchartmaker Pro provides the graph of Q Tests for Fit, which shows the adequacy of the edf for the L, M and S curves by testing whether the Z-scores are normally distributed independent of age. The procedures involved are as follows: The age range is divided into G contiguous groups of approximately equal size. We consider G over a range of values. The Range of G is depending on the sample size n such that max{ n 0 . 25 , max( edf L , edf M , edf S ) + 3} < G < min{ n 0 . 4 , n / 50 } , n > 50 which ensures that v = G - edf > 3 and size of each group g . Calculate Chi statistic of location, scale, skewness and kurtosis for M, S, L and kurtosis for each number of G. Transform the above Chi statistic values to Q statistic. Plot the obtained Q statistic values against corresponding degrees of freedom n, which are expected to be between 1.96 and 1.96 of 95% confidence interval. See Pan and Tim for the details of Q Tests procedures applied in LMSchartmaker. Note LMS method does not compensate for kurtosis.
12.4
edf S Suppose the age range is divided into G contiguous groups of approximately equal size. Let edf M , and edf L
denote the e.d.f. of the fitted M, S and L curves respectively, n denote the total sample size,
ng
Glossary of Statistics
v d2 u and g the group mean and variance, and g and g the group normal equivalent deviates for skewness and kurtosis calculated as described by D'Agostino et al. They are as follows:
zg
Location:
42
H 0 : z g = 0 g = 1,2,...,G,
G
QM =
n g z g2
g =1
degrees of freedom.
Scale:
2 H 0 : d g = 1, g = 1, 2,..., G ,
2/3 2 d g - 1 9( n g - 1) G QS = 2 g =1 9( n g - 1)
v S = G - ( edf
Skewness:
2 dg
+ 1) / 2
degrees of freedom.
H 0 : u g ~ N ( 0 ,1), g = 1,2,..., G ,
G 2 QL = u g g =1
2 v = G - edf which has a c distribution on L
degrees of freedom.
Kurtosis:
H 0 : v g ~ N ( 0 ,1), g = 1,2,..., G ,
G 2 QK = v g g =1
2 which has a c distribution on v K = G degrees of freedom.
12.5
ZQ
2 Q 3 - 1 - v 9v = 2 9v
43
12.6
13
13.1
Acknowledgement
Acknowledgement
Peter Green, for development of the original program. Tam Fry of the Child Growth Foundation for financial support. Harvey Goldstein for encouragement.
14
14.1
Statistical References
References
LMS method Cole, T. J. and Green, P. J. 'Smoothing reference centile curves: The LMS method and penalized likelihood', Statistics in Medicine, 11, 1305-1319 (1992). Cole, T. J., Freeman, J. V. and Preece, M. A. 'British 1990 growth reference centiles for weight, height, body mass index and head circumference fitted by maximum penalized likelihood', Statistics in Medicine, 17, 407-429 (1998). Freeman, J. V., Cole, T. J., Chinn, S., Jones, P. R. M., White, E. M. and Preece, M. A. 'Cross sectional stature and weight reference curves for the UK, 1990', Archives of Disease in Childhood, 73, 17-24 (1995) Q test D'Agostino, R. B., Belanger, A. and D'Agostino, Jr. R. B. 'A suggestion for using powerful and informative tests of normality', The American Statistician, Vol 44, 316-321 (1990). Royston, P. and Wright, E M. 'Goodness-of-fit statistics for age-specific reference intervals', Statistics in Medicine, 19, 2943-2962 (2000). Wilson, E. B. and Hilferty, M. M. 'The distribution of chi-square', Proceedings of the National Academy of Sciences, Washington, 17, 684-688 (1931). Pan, H. & Cole, T. J. 'A comparison of goodness of fit tests for age-related reference ranges', Statistics in Medicine, 23, 1749-1765 ( 2004) Detrended Q-Q plot van Buuren, S. and Fredriks, M. 'Worm plot: a simple diagnostic device for modeling growth reference curves', Statistics in Medicine, 20, 1259-1277 (2001).
44
15