Professional Documents
Culture Documents
Advances in Growth Curve and Structural Equation Modeling Topics From The Indian Statistical Institute Proceedings 2017 Ratan Dasgupta
Advances in Growth Curve and Structural Equation Modeling Topics From The Indian Statistical Institute Proceedings 2017 Ratan Dasgupta
https://textbookfull.com/product/multilevel-structural-equation-
modeling-bruno-castanho-silva/
https://textbookfull.com/product/proceedings-of-
secon19-structural-engineering-and-construction-management-
kaustubh-dasgupta/
https://textbookfull.com/product/proceedings-of-
secon-2020-structural-engineering-and-construction-management-
kaustubh-dasgupta/
Multiple Regression and Beyond An Introduction to
Multiple Regression and Structural Equation Modeling
Timothy Z. Keith
https://textbookfull.com/product/multiple-regression-and-beyond-
an-introduction-to-multiple-regression-and-structural-equation-
modeling-timothy-z-keith/
https://textbookfull.com/product/corporate-governance-and-
contingency-theory-a-structural-equation-modeling-approach-and-
accounting-risk-implications-1st-edition-abdul-ghofar/
https://textbookfull.com/product/a-primer-on-partial-least-
squares-structural-equation-modeling-pls-sem-2nd-edition-joseph-
f-hair-jr/
Advances in
Growth Curve and
Structural Equation
Modeling
Topics from the Indian Statistical
Institute—Proceedings 2017
Advances in Growth Curve and Structural Equation
Modeling
Ratan Dasgupta
Editor
123
Editor
Ratan Dasgupta
Theoretical Statistics
and Mathematics Unit
Indian Statistical Institute
Kolkata, West Bengal, India
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
v
Contents
vii
viii Contents
Contributors
ix
x Editor and Contributors
Ratan Dasgupta
R. Dasgupta (B)
Theoretical Statistics and Mathematics Unit, Indian Statistical Institute,
203 B.T. Road, Kolkata 700108, India
e-mail: rdgupta@isical.ac.in; ratandasgupta@gmail.com
fairly accurately, but they do not think naturally in it; instead they tend to revert to
the less efficient but more deeply rooted mother tongue.
The individuals are interviewed at different places of congress like students in
different schools, marketplaces, church, and villages after stratification. Selection of
individuals for interview is made by simple random sampling without replacement
at final stage. Non-response of a unit is remedied by selecting an adjacent available
unit.
In this pilot survey, data are collected in the way of personal interviews, admin-
istering tests and filling up the questionnaire by selected individuals. Sometimes
resistance to fill up the questionnaire was faced from skeptic elements, motif and
ultimate goal of such studies were questioned and monetary angle out of the results
in conducted survey was prodded into, while working in some interior parts of the
state. It appears that affinity of inhabitants of interior hilly region and plane region
where modern facilities are available is slowly improving over time with progress of
education and lifestyle.
In the next section, we present the results obtained from the data analyzed. The
error components in the scores may not be symmetric in all cases.
The individuals may not be highly educated in general, but their receptivity of
to new concepts is commendable. Figures based on scores provide insight into the
lifestyle and educational status of individuals interviewed.
Fluctuations in scores are modeled by martingale differences. Excessive deviation
results for martingale are obtained.
In Sect. 2, we present the results in brief. These are further elaborated in figures
with legends and associated explanation given in detail therein. Section 3 discusses
the results obtained. Fluctuations of observations around the response curves are
explained by martingale theory. While estimating the response curve by nonparamet-
ric regression, assume that the differences of the existing data point from regression
curve behave like martingale differences as new data are gathered from recent inter-
view to modify the regression, and those successive martingale differences constitute
a martingale. Maximal fluctuation of the martingales is of interest.
Excessive deviation results for maximal fluctuations are proved in the Appendix.
Under assumptions on moment bounds, those are not so stringent on martingale dif-
ferences, and deviation probabilities of maximal fluctuations in martingales are seen
to decay with order higher than any polynomial power of the boundary value. Since
the total score and the component scores on IQ and Math aptitude are standardized
in the range [0, 100], we may assume that the martingale differences are bounded.
In such a situation, for the martingales {Mi : 1 ≤ i ≤ n}, we compute the exces-
λ2
sive deviation probability as P(max1≤i≤n |Mi | ≥ λn 1/2 ) = O(e− 2 (1+o(1)) ); λ →
∞, which is similar to the tail probability of normal distribution. High oscillations
of observations from the response curve may be assessed in terms of normal deviate
to identify assignable causes.
In general, the level of IQ is seen to be higher than that of scores in mathematical
aptitude. A large segment of the individuals under the scope of study understood the
4 R. Dasgupta
Fig. 1 Longitudinal score (Math plus IQ) of 47 tribal individuals in Tripura. Longitudinal growth
curves of total scores for 47 tribal individuals in Mathematics and IQ after scaling are shown in
Fig. 1, the maximum possible score is scaled to 100. The number of Math questions is 15, and the
number of IQ questions is 7 with a total of 22 questions administered in interviews. Indications of
upward trend in second interview scores are possibly due to fruitful interaction with the individuals
at the first interview date, on topics at introductory level, before these tests were administered.
Elaboration of the subject matter affects different individuals over a considerable period of time,
and this phenomenon is apparent in observed peaks. Too many lines make the picture congested.
Some individuals appeared only twice, which results in two interview scores over time
aim of the tests and were cooperative with the interviewer on successive occasions
of test administration.
2 Results
The survey conducted over a time span of about 3 years revealed that IQ test score in
percentage is at a higher level than the Math score percentage. Data analyzed under
the assumption that error component is symmetrically distributed and is consistent
with other nonparametric methods to estimate the growth curve for total score of
Math plus IQ, and Math score alone. Symmetry of error component may not hold for
IQ score, expressed in percentage. An upward trend is seen on scores after interaction
with the interviewer, indicating receptivity of individuals towards improvement of
knowledge. Error component in IQ score may not be symmetric. From the growth
curve of Math aptitude and IQ scores, it appears that the education status of tribal
individuals is improving over time. Some of the scores seem to attain stability over
time towards the far end. Different features of collected data and growth curve related
results are further explained in the Figures.
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 5
Fig. 2 Longitudinal score (IQ) of 47 tribal individuals. Questions on Math and IQ are separated and
scores are also separated into two components and expressed in percentage. The first component
IQ score in percentage for longitudinal study is shown in Fig. 2. Patterns in these growth curves
are similar to that of Fig. 1, the level of IQ scores has relatively higher elevation than total score
expressed in percentage
Fig. 3 Longitudinal score (Math) of 47 tribal individuals. The second component, Math score in
longitudinal study is shown in Fig. 3. Elevation of scores is a bit lower that earlier two graphs, and
pattern of the graph is the same with former two graphs. Overall performance in Mathematics seems
to be poor compared to that in IQ
3 Discussions
Fig. 4 Band of growth curves (Math plus IQ) of 47 individuals. The band that contains all 47
longitudinal curves on total score is shown by joining the extreme points with straight lines. For
each value of time x, the line of y coordinate has shortest length within the band containing all
individual curves. The curve joining mid points of the upper and lower band is also shown in
Fig. 4. The curve in the middle represents an estimate of overall growth in total scores under the
assumption that error components, i.e., deviation of scores from response curve for each time point
t have symmetric distribution. Mid-range is sometimes more efficient than the mean for estimating
central tendency in a number of distributions with limited range, see, e.g., Rider (1957)
Fig. 5 Lowess on mid points of growth band (Math + IQ) of 47 individuals. A nonparametric
growth curve of total score by lowess regression on mid points of the band with f = 2/3 is shown
in Fig. 5. The curve has initial upward trend for a while, it goes down and reaches a minimum
when initial boost from interaction with interviewer possibly erodes. Then the curve shows an
upward trend gradually over time. We shall see later that the pattern of the growth curve remains
the same, if the median of the scores on days on recording is considered instead of midpoints of the
band considered in lowess regression. This is in concordance with the assumption that errors have
symmetric distribution for total score
curve may then be compared with normal tail probability to investigate for assignable
causes.
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 7
Fig. 7 Lowess on mid points of growth band (IQ) of 47 individuals. A nonparametric growth curve
of IQ score by lowess regression on midpoints of the band with f = 2/3 is shown in Fig. 7. The
curve has initial downward trend and reaches a minimum, then the curve gradually exhibits an
upward trend over time. We shall see later in Fig. 26 that the pattern of the graph changes when
the assumption of symmetric error is not made, indicating that the assumption of symmetric error
component for IQ scores may not hold
Fig. 8 Band of growth curves (Math) of 47 individuals. The band containing 47 longitudinal curves
on mathematics score is shown in Fig. 8. The middle curve joining the midpoints of the band provides
an estimate of growth curve in mathematics score
8 R. Dasgupta
Fig. 10 Proliferation rate of total score: wt. exp(−x); spline. We compute the proliferation rate
d
based on the growth curve shown in Fig. 5. Proliferation rate dt log y = 1y dy
dt is a scaled version of
velocity dy
dt . This measure is independent of the choice of unit used in measuring y. For total score
y with growth curve computed in Fig. 5 by lowess regression, the proliferation rate is obtained in
Fig. 10. The curve has an initial downward trend up to day 190, and then the curve rises gradually,
with fast growth towards the end. Computation is based on a technique proposed in Dasgupta
(2015), with exponentially decaying weights attached to empirical slopes computed from data pairs
at different time points with respect to a fixed time point t of interest. More weights are given to
data points near the time t of derivative computation, and less weights to distant time points from t.
Weighted mean of these empirical slopes at derivative stage and smooth.spline with spar = 0.01 at
smoothing stage in SPlus provide proliferation rate at time point t, when divided by y. Instead of
considering median or trimmed mean that produces too smooth rates especially towards end of the
range, we take weighted average of empirical slopes; with less weight assigned to slopes computed
from distant points
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 9
0.0010
0.0005
Proliferation rate of score/day
0.0
-0.0005
-0.0010
d
Fig. 11 Proliferation rate of IQ score: wt. exp(−x); spline. Proliferation rate dt log y over 11 time
points for IQ score y is computed as in Fig. 10, with exponentially decaying weights attached to
empirical slopes computed from data pairs at different time points with respect to a fixed time
point t of interest. Weighted mean of the empirical slopes is considered at derivative stage of
computation and the package smooth.spline with spar = 0.01 at smoothing stage in SPlus provides
the proliferation rate at the time point t. A consistent sharp rise in the beginning and then slightly
slow rise of rate towards the end is seen in Fig. 11. However, we shall see later that the pattern
changes when we consider median instead of mid-values of the band to estimate central tendency.
This is because in Fig. 7, the basic input shown for computing proliferation rate here changes to a
different pattern when the assumption of symmetric error for IQ is dropped
0.0010
Proliferation rate of score/day
0.0005
0.0
-0.0005
-0.0010
d
Fig. 12 Proliferation rate of math score: wt. exp(−x); spline. Proliferation rate dt log y over
11 time points for Math score y is computed as in Fig. 10, with exponentially decaying weights
attached to empirical slopes computed from data pairs at different time points with respect to a fixed
time point t of interest. Weighted mean of the empirical slopes is considered at derivative stage
of computation and the package smooth.spline with spar = 0.0001 at smoothing stage in SPlus
provide the proliferation rate at the time point t. A downward trend in the beginning and then a
typical sharp rise towards middle is observed in Fig. 12; the rate slows down towards the end and
approaches stability over time
10 R. Dasgupta
Fig. 13 Longitudinal score (Math plus IQ) of 19 individuals with more than two recordings.
Analysis based on longitudinal data with many recordings per individuals has accurate assertions.
We now consider only those cases where more observations are available per individuals, as there are
some longitudinal growth curves with only two recordings. When tracked over time, 19 individuals
had more than two recordings. Total scores of Math plus IQ for these individuals are shown in
Fig. 13. Figure 13 is less messy compared to Fig. 1. Upward trend of scores in general are seen after
the first interview
Fig. 14 Longitudinal score (IQ) of 18 individuals with more than two recordings. Scores on IQ for
18 individuals having more than two recordings and nonzero score are shown in Fig. 14. The scores
are of higher elevation compared to that for total score of previous picture. Upward trend of scores
is more prominent in the second interview, compared to the previous figure
Appendix
To study the deviation of the observed growth from nonparametric response curve
derived under mild assumption and to have a bound on maximum fluctuation of
deviations, consider these errors to be martingale differences. An assignable cause
may be looked into, if the fluctuations are of different type other than specified by
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 11
Fig. 15 Longitudinal score (Math) of 19 individuals with more than two recordings. Scores on
Math for 19 individuals with more than two recordings are shown in Fig. 15. The scores are of
lower elevation compared to those for total score and IQ scores of previous pictures. Upward trend
of scores is prominent in general immediately after the first interview
Fig. 16 Band of growth curves (Math + IQ) of 19 Individuals with >2 recordings. The band
containing 19 longitudinal curves on total score is shown. The curve joining mid points of the upper
and lower band is also shown in Fig. 16. The curve in the middle represents an estimate of overall
growth in total scores under the assumption that errors are symmetric. With less number of curves,
the picture is less messy. An individual scoring zero in IQ, scored a positive mark in Math. Thus,
the number of individuals with nonzero total score is 19 here
Fig. 17 Lowess on mid points of growth band (Math + IQ) of 19 individuals with >2 recordings.
A nonparametric growth curve of total score by lowess regression on mid points of the band with
f = 2/3 is shown in Fig. 17. The curve goes down over time towards stability
12 R. Dasgupta
Fig. 18 Band of growth curves (IQ) of 19 individuals with >2 recordings. A similar band containing
of 18 longitudinal curves on IQ score is shown in Fig. 18. Here again, the curve in the middle
represents an estimate of overall growth of IQ score when errors have symmetric distribution. The
pattern of growth is clearer with less number of individuals
Fig. 19 Lowess on mid points of growth band (IQ) of 18 Individuals with >2 recordings. A
nonparametric growth curve of IQ score by lowess regression on midpoints of the band with f = 2/3
is shown in Fig. 19. The curve has slight upward trend initially after a drop, and gradually lowers
over time. This feature of the graph may be due to the assumption made on symmetric error
Fig. 20 Band of growth curves (Math) of 19 individuals with >2 recordings. The band containing
19 longitudinal curves on Math score is shown in Fig. 20. The curve in the middle represents an
estimate of overall growth of Math score. The oscillatory pattern of growth has slight upward trend.
This becomes clear with less number of individuals
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 13
Fig. 21 Lowess on midpoints of growth band (Math) of 19 individuals with >2 recordings. A
nonparametric growth curve of Math score by lowess regression on midpoints of the band with
f = 2/3 is shown in Fig. 21. The curve indicates an upward trend, after slowing down slightly at
the initial stage
-0.001
Proliferation rate of score/day
-0.002
-0.003
-0.004
Fig. 22 Proliferation rate of 19 persons total score: wt. exp(−x); spline. We compute the prolifer-
d
ation rate based on the growth curve shown in Fig. 17. Proliferation rate dt log y over 9 time points
for total score y is computed based on a technique proposed in Dasgupta (2015), with exponentially
decaying weights attached to empirical slopes computed from data pairs at different time points
with respect to a fixed time point t of interest. Weighted mean of the empirical slopes is considered
at derivative stage of computation and the package smooth.spline with spar = 0.01 at smoothing
stage in SPlus provides the proliferation rate at the time point t. The curve in Fig. 22 has a sharp rise
in the beginning and this gradually slows down towards the end. With less number of 19 individuals,
the curve is different compared to that with 47 individuals shown in Fig. 10
-0.0020
Proliferation rate of score/day
-0.0028 -0.0026 -0.0024 -0.0022
d
Fig. 23 Proliferation rate of 18 persons IQ score: wt. exp(−x); spline. Proliferation rate dt log y
over 9 time points for IQ score y is computed like Fig. 10, with exponentially decaying weights
attached to empirical slopes computed from data pairs at different time points with respect to a fixed
time point t of interest in Fig. 23. Weighted mean of the empirical slopes is considered at derivative
stage of computation and the package smooth.spline with spar = 0.01 at smoothing stage in SPlus
provides the proliferation rate at the time point t. A sharp rise of proliferation rate is seen in the
beginning, which gradually slows down and then a downward trend of proliferation rate is seen.
The rate of fall is sharper towards the end
Proliferation rate of score/day
-0.0010
-0.0020
-0.0030
d
Fig. 24 Proliferation rate of 19 persons Math score: wt. exp(−x); spline. Proliferation rate dt log y
over 9 time points for Math score y is computed based on a technique proposed in Dasgupta (2015),
with exponentially decaying weights attached to empirical slopes computed from data pairs at
different time points with respect to a fixed time point t of interest in Fig. 24. Weighted mean of the
empirical slopes is considered at derivative stage of computation and the package smooth.spline
with spar = 0.0001 at smoothing stage in SPlus provides the proliferation rate at the time point t. A
little bit downward trend is seen in the beginning reaching a minimum near 80 days, then the curve
gradually moves up fast to reach a high value at time t = 700 days. The curve reaches stability at
the far end. Rise of proliferation rate is slow towards the end
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 15
Fig. 25 Lowess on median of growth band (Math + IQ) of 47 individuals. We now examine the
data with assumption of symmetric error dropped. Instead, we consider median to calculate the
response curve. A nonparametric growth curve of total score by lowess regression on a robust
estimate of central tendency as median scores at time points fixed on dates of interview in the
band, with f = 2/3 is shown in Fig. 25. The curve dispenses off the stringent assumption that error
components are symmetric. The curve has an initial upward trend for a while, then it goes down
and reaches a minimum when initial boost from interaction with interviewer erodes; then the curve
gradually shows an upward trend over time
Fig. 26 Lowess on median of growth band (IQ) of 47 individuals. Growth curve of IQ score by
lowess regression on median of scores at time points fixed on dates of interview in the band, with
f = 2/3 is shown. Elevation of this curve is a bit higher than the previous graph on total score.
Like in the previous curve, this has initial upward trend for a while, then it goes down and reaches
a minimum when initial boost from interaction with interviewer possibly erodes; then the curve
gradually shows an upward trend over time. The pattern of the curve in Fig. 26 is not similar to that
shown in Fig. 7, indicating that the assumption of symmetric error component for IQ scores may
not hold
Fig. 27 Lowess on median of growth band (Math) of 47 individuals. Growth curve of Math score
by lowess regression on median of scores at time points fixed as dates of interview in the band, with
f = 2/3 is shown in Fig. 27. Elevation of this curve is lower than the previous two graphs. Like
the previous curve, this has initial upward trend for a while. Then it goes down further compared to
Fig. 9 and reaches a minimum when initial boost from interaction with interviewer possibly erodes;
then the curve gradually shows an upward trend over time. The pattern is similar to that for Fig. 9
0.001
Proliferation rate of score/day
0.0
-0.001
-0.002
-0.003
-0.004
Fig. 28 Proliferation rate of 47 persons total score (med): wt. exp(−x); spline. Proliferation rate
d
dt log y over 11 time points for total score y is computed like Fig. 10, with exponentially decaying
weights attached to empirical slopes computed from data pairs at different time points with respect
to a fixed time point t of interest in Fig. 28. Weighted mean of the empirical slopes is considered at
derivative stage of computation and the package smooth.spline with spar = 0.0001 at smoothing
stage in SPlus provides the proliferation rate at the time point t. A downward trend in the beginning
is seen and then a typical sharp rise is seen to reach a maximum at t = 700 days, the peak value
slightly crossing the initial value. A bit downward trend is seen towards the end to reach stability
of proliferation rate
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 17
0.002
Proliferation rate of score/day
0.001
0.0
-0.003 -0.002 -0.001
Fig. 29 Proliferation rate of 47 persons IQ score (med): wt. exp(−x); spline. Proliferation rate
d
dt log y over 11 time points for IQ score y is computed based on a technique proposed in Dasgupta
(2015), with exponentially decaying weights attached to empirical slopes computed from data pairs
at different time points with respect to a fixed time point t of interest in Fig. 29. Weighted mean of
the empirical slopes is considered at derivative stage of computation and the package smooth.spline
with spar = 0.0001 at smoothing stage in SPlus provide the proliferation rate at the time point t. A
steep fall in the beginning is seen to reach a minimum at t = 250 days and then a rise upto t = 700
days is seen with peak value less than the initial value of proliferation rate seen at start. Finally
stability at the far end is seen
0.002
Proliferation rate of score/day
0.0
-0.002
-0.004
Fig. 30 Proliferation rate of 47 persons Math score (med): wt. exp(−x); spline. Proliferation rate
d
dt log y over 11 time points for Math score y is computed like Fig. 10, with exponentially decaying
weights attached to empirical slopes computed from data pairs at different time points with respect
to a fixed time point t of interest in Fig. 30. Weighted mean of the empirical slopes is considered at
derivative stage of computation and the package smooth.spline with spar = 0.0001 at smoothing
stage in SPlus provides the proliferation rate at the time point t. A sharp fall in the beginning is
seen reaching a minimum at t = 250 days and then a sharp rise of the curve is seen, with rate
approximately 0.002 at t = 700 days, to cross the initial value of rate. Ability to learn mathematics
is commendable in individuals. Finally, the proliferation rate shows a downward trend to reach
stability at the far end
18 R. Dasgupta
Fig. 31 Lowess on mean of growth curves (Math + IQ) of 47 individuals. Although not robust,
mean is a natural estimate in sample for central tendency. Instead of median, we consider lowess
with f = 2/3 on mean of growth curves in Fig. 31. Rise of the curve for total score is seen at the
beginning; it then reaches a minimum and gradually rises again till the end
Fig. 32 Lowess on mean of growth curves (IQ) of 47 individuals. In this lowess curve with f = 2/3,
the fall from high initial value is steep. Then the curve in Fig. 32 rises again towards a high value
of 66.95%
Fig. 33 Lowess on mean of growth curves (Math) of 47 individuals. In Fig. 33, unlike Fig. 32 the
fall is not that steep in this lowess curve with f = 2/3 for Math score. After a fall, the curve rises
again to reach a value of 47.80%. This is lower than the corresponding value 66.95% in IQ, shown
in the previous figure
Longitudinal Studies on Mathematical Aptitude and Intelligence Quotient … 19
0.001
Proliferation rate of score/day
0.0
-0.001
-0.002
-0.003
Fig. 34 Proliferation rate of 47 persons total score (mean): wt. exp(−x); spline. The proliferation
rate curve in Fig. 34 with spar = 0.0001 mimics the pattern of Fig. 28 where median is considered in
place of mean. The patterns are very similar in either case of mean or median being used to estimate
the growth curve of total score, the similarity of Figs. 27 and 31 is reflected in the similarity of
proliferation rate curves computed under different assumptions
An extra term 2 in cν above appears due to the fact that expectation of max-
imum of the terms in (2.2) of Dasgupta (1993) is bounded above by sum of the
expectations in (2.3) therein, leading to an extra factor 2; i.e., the correct expression
is E max(|Sn |ν−2 X n2 , |Sn∗ |ν−2 X n2 ) < E(|Sn |ν−2 X n2 + |Sn∗ |ν−2 X n2 ). The bound is useful
in estimating remainder from the main part in nonlinear statistics that arise in many
situations related to limit theorems, see, e.g., Dasgupta (1994). The modification
does not affect the results of Dasgupta (1994) as L is a generic positive constant
therein.
Consider Mi = ij=1 y j , where M is a martingale and y s are the martingale
differences. For a martingale M with finite νth moment, one may write From Doob’s
inequality
20 R. Dasgupta
0.0010
Proliferation rate of score/day
0.0005
0.0
-0.0005
-0.0010
Fig. 36 Proliferation rate of 47 persons math score (mean): wt. exp(−x); spline. Proliferation rate
with spar = 0.0001 in Fig. 36 with mean values for Math scores has a sharp downward trend in the
beginning. Then the curve rises again to achieve a peak with lower value than the initial value at
start, before stability is achieved; this pattern is unlike Fig. 30 with median score, where the peak is
higher than the initial value
Next consider the case where all the moments of martingale differences exist, but
the moment generating function need not necessarily exist. Consider the following
moment bounds for the martingale differences y j .
Type 1:
α
sup E|y j |ν ≤ Lew0 ν (4)
j≥1
∀ν > 1, and for some L > 1, where w0 > 0, α > 1. In what follows L(> 1) is a
generic constant. The above condition is equivalent to
Another random document with
no related content on Scribd:
Music for the New Church
As the kingdom changed its king at the death of each monarch, the
country swayed from Catholicism to Protestantism and back again,
and many a poet and musician lost his head or was burnt at the stake
because he wrote for the Protestant Church. In the case of Marbeck
who had made music for the Book of Common Prayer, he just
escaped death for the crime of writing a Bible concordance (an
index)!
Before Wynken de Worde’s song book came out, William Caxton,
the great printer, published a book called Polychronicon by Higden.
In this, was an account of Pythagoras and his discovery of tone
relations (Chapter IV); this proves the great interest in England for
the science, as well as the art, of music.
In Frederick J. Crowest’s book, The Story of the Art of Music, he
tells very simply the state of music in England at this time:
“When the adventurous Henry VIII plunged into and
consummated (completed) the reformation scheme, it was at the
expense of considerable inconvenience to musicians obliged,
perforce, to change their musical manners as well as their faith. In
double quick time the old ecclesiastical (church) music had to be cast
aside, and new church music substituted.... This meant pangs and
hardships to the musicians, possibly not too industrious, accustomed
to the old state of things. Simplicity, too, was the order, a change that
must have made musicians shudder when they, like others before
them, from the time of Okeghem, had regarded the Mass as the
natural and orthodox (correct) vehicle for the display of the
contrapuntal miracles they wrought.”
Now the Mass became the “Service,” and the motet was turned
into the “Anthem,” which we still use in our churches. Most of the
famous composers of the 16th and 17th centuries in England wrote
for the new Anglican or Protestant Church, and made the new music
lovely indeed. Many of them were organists or singers in the Chapel
Royal, so they had been well prepared for their work.
To make this new music different from the old, the writers were
ordered to fit every syllable with a chord (in the harmonic style). In
the old counterpoint, of course, the words were somewhat blurred.
These experiments with chords did much to free music for all time.
One of the earliest of the church composers is Thomas Tallis
(about 1520–1585), a “Gentleman of the Chapel Royal” and father of
English cathedral music. Through his long career, Tallis followed the
different religions of the rulers from Henry VIII to Elizabeth, writing
Catholic music or Protestant as was needed. You see he liked his
head, so he changed his music with each new monarch. He, like some
of the composers of the Netherlands school, wrote a motet for forty
voices.
He shared with his pupil, William Byrd, the post of organist of the
Chapel Royal, and together they opened a shop “to print and sell
music, also to rule, print and sell music paper for twenty-one years”
under a patent granted by Queen Elizabeth to them only. How
successful the two composers were in the publishing business is not
stated, but at least they could publish as many of their own works as
they cared to! After Tallis’ death, in 1585, for a while Byrd ran the
shop alone, and published a collection of Psalms, Sonets, and Songs
of Sadness and Pietie. In this was written “Eight reasons briefly set
down by the Author (Byrd) to persuade every one to learn to sing” to
which he added:
Since singing is so goode a thing
I wish all men would learne to sing.
Famous Old Music Collections
England was the land of famous music collections in the 16th and
17th centuries. The first of these by Byrd was a book of Italian
Madrigals with English words, Musica Transalpina, (Music from
across the Alps). The entire title was (Don’t laugh!): “Musica
Transalpina; Madrigals translated of foure, five, and sixe parts,
chosen out of diuers excellent Authors, with the first and second part
of La Virginella, made by Maister Byrd vpon two Stanz’s of Ariosto
and brought to speak English with the rest. Published by N. Yonge, in
fauer of such as take pleasure in Musick of voices. Imprinted at
London by Thomas Easy, the assigne of William Byrd, 1588. Cum
Privilegio Regise Maiestatis (With permission of her Royal
Majesty).” A long title and one that would not make a book a
“bestseller” today! Do notice how they mixed u’s and v’s and put in
e’s where you least expect them!
There were fifty-seven madrigals in the long titled collection
including the two by “Maister Byrd”; the others were by the Italian
and Netherland madrigal writers, such as Palestrina, Orlandus
Lassus and Ferabosco, a composer of masques and madrigals, who
lived for years in England.
Byrd’s compositions in this work mark the beginning of the great
English school of madrigals, which were so lovely that this period
(1560 to 1650) was called the “Golden Age.”
The Golden Age of Madrigals
Another most valuable collection was for many years called Queen
Elizabeth’s Virginal Book, but is now the Fitzwilliam Virginal Book,
and the original manuscript is in the Fitzwilliam museum at
Cambridge. It was supposed to have been “Good Queen Bess’” book,
but it was not, as some of its compositions were composed after her
death. It is not known who copied 220 pages of music, but it may
have been a wealthy Roman Catholic, Francis Tregian, who spent
twenty-four years in prison on account of his religious faith. This
name, abbreviated or in initials, is found in several places in the
manuscript. An edition in our notation has been made by J. A. Fuller
Maitland and W. Barclay Squire. Many of the old songs of English
minstrelsy are found among the numbers, and they were arranged
for the instrument by the famous composers of that day. There are
also original compositions as well as “ayres” and variations. Among
the composers we find Dr. John Bull, Thomas Morley, William Byrd,
Orlando Gibbons, Giles Farnaby, Richard Farnaby, Thomas Tallis,
Jan Sweelinck, the Dutch organist, and many others. Here are some
of the quaint titles: St. Thomas’ Wake, King’s Hunt, The Carman’s
Whistle, The Hunt’s Up, Sellonger’s Round, Fortune My Foe, Pawles
Warfe, Go from My Window, Bonny Sweet Robin, besides many
Pavanes, Galiardes, Fantasias, and Preludiums.
There is also a collection of Byrd’s virginal music called My Lady
Nevell’s Booke. Lady Nevell may have been a pupil of Byrd. There are
two collections of this same kind at Buckingham Palace, the home of
the King of England,—Will Forster’s Virginal Book and Benjamin
Cosyn’s Virginal Book. In the index of the latter, we read: “A table of
these Lessons following made and sett forth by Ben Cos.” In all, he
copied more than 90 compositions!
Later came John Playford, music publisher, whose first musical
publication, The English Dancing-Master (1650), contains almost a
hundred old folk tunes. Select Musical Ayres appeared three years
later, and is a typical 17th century song collection of first-class poems
by Jonson, Dryden and others set to music by well-known
composers. His book on the theory of music, used for almost a
century, contained “lessons” for the viol, the cithern and flageolet.
His Dancing Master, a collection of airs for violin for country
dances, has brought to us many popular ballad tunes and dance airs
of the period.
In these collections we often find the names Fancies, Fantazia, or
Fantasies, a type of composition that grew out of the madrigal and
led to the sonata. It was the name given to the first compositions for
instruments alone like the ricercari of the Italians, which were
original compositions and not written on a given subject (called in
England “ground”), or on a folk song. The Fancies were sometimes
written for the virginal, and sometimes for groups of instruments
such as a “chest of viols” or even five cornets(!).
The Chest of Viols
“Chest of Viols” may sound queer to you, but it isn’t! It was the
custom in England at that time for people to have collections of
instruments in or out of chests. So, when callers came they could
play the viol, instead, probably of bridge! You can read about these
interesting old days in Samuel Pepys’ Diary. He played the lute, the
viol, the theorbo, the flageolet, the recorder (a kind of flute) and the
virginal, and he was the proud owner of a chest of viols. He always
carried his little flageolet with him in his pocket, and he says that
while he was waiting in a tavern for a dish of poached eggs, he played
his flageolet, also that he remained in the garden late playing the
flageolet in the moonlight. (Poetic Pepys!)
Thomas Morley, Byrd’s pupil, who was made a partner in the
publishing house after Tallis’s death, wrote his madrigals for virginal,
and a collection called First Book of Consort Lessons for Six
Instruments, Lute, Pandora, Cittern (an old English form of guitar),
Bass Viol, Flute, and Treble Viol, and much sacred music. He also
wrote a Plaine and Easie Introduction to Practical Musick, a book of
great value and interest to musicians for the last three centuries, for
it is a mirror of his time and of his fellow composers.
He tells of a gentleman, who, after dinner, was asked by his
hostess to sing from the music she gave him. It was the custom in
England to bring out the music books after dinner and for the guests
to play and sing, as we wind up our graphophones and switch on the
radio. The gentleman stammeringly declared that he could not sing
at sight and “everyone began to wonder; yea, some whispered to
others, demanding how he was brought up.” He was so ashamed of
his ignorance that he immediately took music lessons to remedy his
woeful lack of culture. This proves that musical education was not
looked upon as a luxury but a necessity in the 17th century.
Truly, it was a musical era, this time of Morley and Byrd! Fancy
playing, while waiting for the barber, the viol, flute, lute, cittern, or
virginal left for that purpose. Yet what would our dentist do today if
he had to listen to a saxophone and jazz chorus from his waiting
room? In those days, too, there was always a bass viol left in a
drawing-room for the guest, to pass the time, waiting for the host to
appear. Think of all the practising you could do waiting for the busy
dentist or eternally late hostess!
The children of people who were poor, were taught music to make
them fit to be “servants, apprentices or husbandmen.” Laneham, a
groom who had been brought up in the royal stable, was advanced to
the post of guarding the door of the council chamber and this is how
he described his qualifications for the job: “Sometimes I foot it with
dancing; now with my gittern, and else my cittern, then at the
virginals (ye know nothing comes amiss to me); then carol I up a
song withal; that by-and-by they come flocking about me like bees to
honey; and ever they cry, ‘Another, good Laneham, another!’” (From
The Story of Minstrelsy by Edmundstoune Duncan.)
Shakespeare and Music
This was the day in which Shakespeare lived, and from his plays
we get a very good idea of the popular music of his time, for he used
bits of folk songs and old ballads. It was a Lover and his Lass from
As You Like It was set to music by Thomas Morley, and is one of the
few songs written to Shakespeare’s words in his own day that has
come down to us. In Twelfth Night there is O Mistress Mine, Hold
thy Peace, Peg-a-Ramsey, O, London is a Fine Town, Three Merry
Men be We, and the Clown’s song:
Hey! Robin, jolly Robin,
Tell me how thy lady does, etc.
The last of the great 17th century English composers, and the
greatest of them all, is Henry Purcell (1658–1695). His father was a
well-known musician, and the uncle, who brought him up, was also a
musician, so the young boy heard much music in his own home, and
no doubt knew many composers.
Sir Frederick Bridge in Twelve Good Musicians tells us that the
Purcell family came from Tipperary in Ireland and that Henry’s
father and uncle were Gentlemen in the Chapel Royal in London.
Henry began his music studies at the age of six, for he, too, was one
of “Captain Cooke’s boys,” and when he was twelve years old,
“Maister Purcell” wrote a composition in honor of “His Majestie’s
Birthday.”
The young Purcell, sometimes called the “English Mozart,” gained
much from Pelham Humphrey who told him of Lully in France. After
Humphrey’s early death (he was only twenty-seven), Purcell studied
with Dr. Blow, and the two musicians were devoted comrades. Their
tombs lie close together near the old entrance of the organ loft,
where they must have spent many hours of their lives.
Matthew Locke was also a friend of Purcell’s, and probably did
much to interest the young composer in the drama, for in spite of his
early church training, Purcell’s greatest offering to English music was
his opera writing. While Purcell’s are not operas in our sense of the
word, they are the nearest thing to them that England had, before the
Italians came with theirs in the 18th century. He wrote music to
masques and plays, several of which were even called operas, yet
only one really was an opera. Purcell’s music “was so far in advance
of anything of the sort known in any part of Europe in his day, in
point of dramatic and musical freedom and scenic quality, that one
can only regret his early death’s preventing his taking to opera
writing on a larger scale.” (W. F. Apthorp.) Among the things he put
to music were the plays of Dryden and of Beaumont and Fletcher.
Purcell was one of the first English composers to use Italian
musical terms, like adagio, presto, largo, etc. He was also one of the
first composers to write compositions of three or four movements for
two violins, ’cello and basso continuo, a part written for harpsichord
or sometimes organ as an accompaniment to the other instruments.
The name of this style of composition also came from the Italian, and
was called Sonata. The first sonatas were composed by Italians. The
word Sonata comes from an Italian word suonare which means to
sound, and was first given to works for instruments. Another form of
composition is the Cantata, from cantare which means to sing. It is a
vocal composition with accompaniment of instruments, a direct
descendant of the motet and madrigal, and of the early oratorios.
The Toccata, too, comes from the Italian toccare, meaning to
touch, and was originally a work for instruments with keyboards.
The Italian language gave us our musical names and terms, because
Italian music was the model of what good music should be, and
England, France and Germany copied Italian ways of composing.
Everyone uses the Italian terms for musical expressions so that all
nationalities can understand them.
When Purcell was only 17 years old, he composed an opera to be
played by young ladies in a boarding school. This was Dido and
Æneas, and it is so good that few writers on musical subjects believe
that it was written in his youth.
In every branch of composition in which Purcell wrote, he excelled.
His church music is the finest of his day, his chamber music and his
operas are looked upon as works of genius. In fact, he is still
considered the most gifted of all English composers.
He was only 37 when he died, and was a very great loss to the
growth of English music.
Music Comes of Age