Experience Seamless Full Ebook Downloads for Every Genre at ebookluna.

com

Longitudinal Data Analysis by Donald Hedeker

https://ebookluna.com/product/longitudinal-data-analysis-by-
donald-hedeker/

OR CLICK BUTTON

DOWNLOAD NOW

Explore and download more ebook at https://ebookluna.com

 Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

(eBook PDF) Longitudinal Data Analysis by Donald Hedeker

https://ebookluna.com/product/ebook-pdf-longitudinal-data-analysis-by-
donald-hedeker/

ebookluna.com

(eBook PDF) Engineering Economic Analysis 13th by Donald

G. Newnan

https://ebookluna.com/product/ebook-pdf-engineering-economic-
analysis-13th-by-donald-g-newnan/

ebookluna.com

(eBook PDF) Engineering Economic Analysis 12th Edition by

Donald G. Newnan

https://ebookluna.com/product/ebook-pdf-engineering-economic-
analysis-12th-edition-by-donald-g-newnan/

ebookluna.com

(Original PDF) Categorical Data Analysis 3rd Edition by

Alan Agresti

https://ebookluna.com/product/original-pdf-categorical-data-
analysis-3rd-edition-by-alan-agresti/

ebookluna.com
(eBook PDF) An Introduction to Categorical Data Analysis
by Alan Agresti

https://ebookluna.com/product/ebook-pdf-an-introduction-to-
categorical-data-analysis-by-alan-agresti/

ebookluna.com

(eBook PDF) Econometric Analysis of Panel Data 5th Edition

by Badi H. Baltagi

https://ebookluna.com/product/ebook-pdf-econometric-analysis-of-panel-
data-5th-edition-by-badi-h-baltagi/

ebookluna.com

(Original PDF) Introduction to Statistics and Data

Analysis 5th Edition by Roxy Peck

https://ebookluna.com/product/original-pdf-introduction-to-statistics-
and-data-analysis-5th-edition-by-roxy-peck/

ebookluna.com

(eBook PDF) Bioinformatics and Data Analysis in

Microbiology

https://ebookluna.com/product/ebook-pdf-bioinformatics-and-data-
analysis-in-microbiology/

ebookluna.com

Data Analysis with Microsoft Power Bi - eBook PDF

https://ebookluna.com/download/data-analysis-with-microsoft-power-bi-
ebook-pdf/

ebookluna.com
 To Vera,Ava, Lina,
and the memory of Daniel
- D.H.

To Carol, Julie, Jason,

Michael, and the memory of
RozEyn, Dorothy, and
Burton
- R.D.G.
This Page Intentionally Left Blank
CONTENTS

Preface xiii

Acknowledgments xvii

Acronyms xix

1 Introduction 1

1.1 Advantages of Longitudinal Studies 1

1.2 Challenges of Longitudinal Data Analysis 2
1.3 Some General Notation 3
1.4 Data Layout 4
1.5 Analysis Considerations 5
1.6 General Approaches 5
1.7 The Simplest Longitudinal Analysis 7
1.7.1 Change Score Analysis 7
1.7.2 Analysis of Covariance of Post-test Scores 8
1.7.3 ANCOVA of Change Scores 9
1.7.4 Example 9
1.8 Summary 12

2 ANOVA Approaches to Longitudinal Data 13

vii
viii CONTENTS

2.1 Single-Sample Repeated Measures ANOVA 14

2.1.1 Design 14
2.1.2 Decomposing the Time Effect 17
2.1.2.1 Trend Analysis-Orthogonal Polynomial Contrasts 18
2.1.2.2 Change Relative to Baseline-Reference Cell Contrasts 19
2.1.2.3 Consecutive Time Comparisons-Profile Contrasts 19
2.1.2.4 Contrasting Each Timepoint to the Mean of Subsequent
Timepoints-Helmert Contrasts 19
2.1.2.5 Contrasting Each Timepoint to the Mean of Others-
Deviation Contrasts 20
2.1.2.6 Multiple Comparisons 20
2.2 Multiple-Sample Repeated Measures ANOVA 21
2.2.1 Testing for Group by Time Interaction 23
2.2.2 Testing for Subject Effect 23
2.2.3 Contrasts for Time Effects 24
2.2.3.1 Orthogonal Polynomial Partition of SS 24
2.2.4 Compound Symmetry and Sphericity 25
2.2.4.1 Sphericity 26
2.3 Illustration 27
2.4 Summary 29

3 MANOVA Approaches to Longitudinal Data 31

3.1 Data Layout for ANOVA versus MANOVA 32
3.2 MANOVA for Repeated Measurements 34
3.2.1 Growth Curve Analysis-Polynomial Representation 34
3.2.2 Extracting Univariate Repeated Measures ANOVA Results 37
3.2.3 Multivariate Test of the Time Effect 38
3.2.4 Tests of Specific Time Elements 38
3.3 MANOVA of Repeated Measures-s Sample Case 39
3.3.1 Extracting Univariate Repeated Measures ANOVA Results 41
3.3.2 Multivariate Tests 41
3.4 Illustration 42
3.5 Summary 45

4 Mixed-Effects Regression Models for Continuous Outcomes 47

4.1 Introduction 47
4.2 A Simple Linear Regression Model 48
4.3 Random Intercept MRM 49
4.3.1 Incomplete Data Across Time 51
4.3.2 Compound Symmetry and Intraclass Correlation 51
4.3.3 Inference 52
 CONTENTS ix

4.3.4 Psychiatric Dataset 52

4.3.5 Random Intercept Model Example 55
4.4 Random Intercept and Trend MRM 56
4.4.1 Random Intercept and Trend Example 58
4.4.2 Coding of Time 59
4.4.2.1 Example 61
4.4.3 Effect of Diagnosis on Time Trends 62
4.5 Matrix Formulation 64
4.5.1 Fit of Variance-Covariance Matrix 66
4.5.2 Model with Time-VaryingCovariates 69
4.5.2.1 Within and Between-Subjects Effects for Time-Varying
Covariates 72
4.5.2.2 Time Interactions with Time-Varying Covariates 74
4.6 Estimation 76
4.6.1 ML Bias in Estimation of Variance Parameters 79
4.7 Summary 79

5 Mixed-Effects Polynomial Regression Models 81

5.1 Introduction 81
5.2 Curvilinear Trend Model 81
5.2.1 Curvilinear Trend Example 83
5.3 Orthogonal Polynomials 86
5.3.1 Model Representations 89
5.3.2 Orthogonal Polynomial Trend Example 90
5.3.3 Translating Parameters 91
5.3.4 Higher-Order Polynomial Models 94
5.3.5 Cubic Trend Example 96
5.4 Summary 99

6 Covariance Pattern Models 101

6.1 Introduction 101
6.2 Covariance Pattern Models 102
6.2.1 Compound Symmetry Structure 102
6.2.2 First-Order Autoregressive Structure 103
6.2.3 Toeplitz or Banded Structure 103
6.2.4 Unstructured Form 104
6.2.5 Random-Effects Structure 104
6.3 Model Selection 105
6.4 Example 105
6.5 Summary 111
X CONTENTS

7 Mixed Regression Models with Autocorrelated Errors 113

7.1 Introduction 113
7.2 MRMs with AC Errors 114
7.2.1 AR( 1) Errors 116
7.2.2 MA(1) Errors 118
7.2.3 ARMA(1,l) Errors 119
7.2.4 Toeplitz Errors 119
7.2.5 Nonstationary AR( 1) Errors 120
7.3 Model Selection 121
7.4 Example 122
7.5 Summary 129

8 Generalized Estimating Equations (GEE) Models 131

8.1 Introduction 131
8.2 Generalized Linear Models (GLMs) 132
8.3 Generalized Estimating Equations (GEE) models 134
8.3.1 Working Correlation Forms 135
8.4 GEE Estimation 136
8.5 Example 138
8.5.1 Generalized Wald Tests for Model Comparison 141
8.5.2 Model Fit of Observed Proportions 144
8.6 Summary 146

9 Mixed-Effects Regression Models for Binary Outcomes 149

9.1 Introduction 149
9.2 Logistic Regression Model 150
9.3 Probit Regression Models 153
9.4 Threshold Concept 154
9.5 Mixed-Effects Logistic Regression Model 155
9.5.1 Intraclass Correlation 158
9.5.2 More General Mixed-Effects Models 158
9.5.3 Heterogeneous Variance Terms 158
9.5.4 Multilevel Representation 160
9.5.5 Response Functions 161
9.6 Estimation 162
9.6.1 Estimation of Random Effects and Probabilities 165
9.6.2 Multiple Random Effects 166
9.6.3 Integration over the Random-Effects Distribution 167
9.7 Illustration 170
9.7.1 Fixed-Effects Logistic Regression Model 173
9.7.2 Random Intercept Logistic Regression Model 175
 CONTENTS xi

9.7.3 Random Intercept and Trend Logistic Regression Model 181

9.8 Summary 186

10 Mixed-Effects Regression Models for Ordinal Outcomes 187

10.1 Introduction 187
10.2 Mixed-Effects Proportional Odds Model 188
10.2.1 Partial Proportional Odds 191
10.2.2 Models with Scaling Terms 194
10.2.2.1 Intraclass Correlation and Partitioning of Between- and
Within-Cluster Variance 195
10.2.3 Survival Analysis Models 196
10.2.3.1 Intraclass Correlation 199
10.2.4 Estimation 199
10.3 Psychiatric Example 202
10.4 Health Services Research Example 212
10.5 Summary 216

11 Mixed-Effects Regression Models for Nominal Data 219

11.1 Mixed-Effects Multinomial Regression Model 220
11.1.1 Intraclass Correlation 222
11.1.2 Parameter Estimation 222
11.2 Health Services Research Example 223
11.3 Competing Risk Survival Models 229
11.3.1 Waiting for Organ Transplantation 229
11.4 Summary 238

12 Mixed-effects Regression Models for Counts 239

12.1 Poisson Regression Model 240
12.2 Modified Poisson Models 24 1
12.3 The ZIP Model 242
12.4 Mixed-Effects Models for Counts 244
12.4.1 Mixed-Effects Poisson Regression Model 244
12.4.2 Estimation of Random Effects 246
12.4.3 Mixed-Effects ZIP Regression Model 247
12.5 Illustration 25 1
12.6 Summary 256

13 Mixed-Effects Regression Models for Three-Level Data 257

13.1 Three-Level Mixed-Effects Linear Regression Model 258
13.1.1 Illustration 260
13.2 Three-Level Mixed-Effects Nonlinear Regression Models 265
xii CONTENTS

13.2.1 Three-Level Mixed-Effects Probit Regression 265

13.2.2 Three-Level Logistic Regression Model for Dichotomous
Outcomes 268
13.2.3 Illustration 268
13.2.4 More General Outcomes 273
13.2.4.1 Ordinal Outcomes 27 5
13.2.4.2 Nominal Outcomes 276
13.2.4.3 Count Outcomes 277
13.3 Summary 27 8

14 Missing Data in Longitudinal Studies 279

14.1 Introduction 279

14.2 Missing Data Mechanisms 280
14.2.1 Missing Completely at Random (MCAR) 28 1
14.2.2 Missing at Random (MAR) 28 1
14.2.3 Missing Not at Random (MNAR) 282
14.3 Models and Missing Data Mechanisms 283
14.3.1 MCAR Simulations 283
14.3.2 MAR and MNAR Simulations 285
14.4 Testing MCAR 289
14.4.1 Example 293
14.5 Models for Nonignorable Missingness 295
14.5.1 Selection Models 295
14.5.1.1 Mixed-Effects Selection Models 296
14.5.1.2 Example 297
14.5.2 Pattern-Mixture Models 302
14.5.2.1 Example 304
14.6 Summary 312

Bibliography 313

Topic Index 335

PREFACE

The prominenceof the longitudinal study has grown tremendously in the past twenty years.
The search for causal inference has led researchers in many fields to collect prospective
longitudinal data in an effort to draw causal links between interventions and endpoints. In
many cases, such studies involve rigorously controlled experiments-for example, prospec-
tive randomized single-center and multi-center clinical trials. In other cases, longitudinal
data from naturalistic studies are either prospectively collected or retrospectively obtained
to explore dynamically changing associations. For example, in a randomized clinical trial,
investigators often collect prospective longitudinal data on one or more endpoints in re-
sponse to a particular intervention relative to a control condition. The focus may be either
to determine if there is a significant difference between control and treated individuals at
the end of the study, often termed an “endpoint” analysis, or to examine differential rates
of change over the course of the study in treated and control conditions. In the case in
which subjects are initially randomized to the control and treatment conditions, differences
in either the final response or the rate of response over time ( e.g., differential linear trends
over time) are taken as evidence that the treatment produces an impact on the outcome mea-
sure of interest above and beyond chance expectations based on responses in the control
condition.
By contrast, in naturalistic studies, subjects who elect to receive a particular treatment
are compared to subjects who do not elect to receive that treatment. Often, the outcome
of interest is repeatedly measured over the course of the study or in the available data.
Furthermore, the timing and intensity of treatment are not controlled, so the treatment or
intervention itself may exhibit longitudinal variability. For example, consider a ten-year
health services research study in which the total amount of mental health service utilization
is the endpoint, and the comparison of interest is between subjects with and without private

xiii
xiv PREFACE

insurance. At the beginning of the study, a subject may not have private insurance, however,
five years into the study, the subject may obtain private insurance. The need to capture
the time-varying nature of the intervention of interest is also a distinguishing feature of
longitudinal studies, The longitudinal nature of the study and the time-varying treatment
variable allow each subject to serve as his or her own control. Of course, in naturalistic
longitudinal studies, we must also question whether those subjects who obtained private
insurance during the course of the study are comparable to those subjects who did not, in
terms of a myriad of other factors that might effect mental health services utilization. For
example, individuals who obtain private insurance may do so because they have a family
history of mental illness and are concerned that public insurance will not meet their mental
health care needs in the future. As such, we would expect that these individuals would have
greater mental health service utilization in general, regardless of whether they did or did
not obtain private insurance. In this example, family history of mental illness and obtaining
private insurance are “confounded:’ and the process by which this confound has arisen is
termed a “selection effect.” Longitudinal studies of this kind are widespread in economics,
epidemiology, sociology, psychology in specific, and social sciences and medical sciences
in general, and tools for reducing or eliminating bias in naturalistic studies have been
studied in considerable detail by Cochran [1968], Heckman [1979], Rubin [1974, 19771,
Rosenbaum and Rubin [1983], Angrist et al. [19961, and Little and Rubin [2000], to name
but a few.
While it would appear that longitudinal studies are now considered foundational for
drawing causal inference, they are not without limitations. Perhaps the most dramatic
difficulty is the presence of missing data. Stated quite simply, not all subjects remain
in the study for the entire length of the study. Reasons for discontinuing the study may
be differentially related to the treatment. For example, some subjects may develop side
effects to an otherwise effective treatment and must discontinue the study. Alternatively,
some subjects might achieve the full benefit of the study early on and discontinue the
study because they feel that their continued participation will provide no added benefit.
The treatment of missing data in longitudinal studies is itself a vast literature, with major
contributions by Laird [1988], Little [1995], Rubin [1976], and Little and Rubin [2002],
to name a few. The basic problem is that even in a randomized and well-controlled
clinical trial, the subjects who were initially enrolled in the study and randomized to the
various treatment conditions may be quite different from those subjects that are available
for analysis at the end of the trial. If subjects “drop out” because they already have derived
full benefit from an effective treatment, an analysis that only considers those subjects who
completed the trial may fail to show that the treatment was beneficial relative to the control
condition. This type of analysis is often termed a “completer” analysis. To avoid this type
of obvious bias, investigators often resort to an “intent to treat” analysis in which the last
available measurement is carried forward to the end of the study as if the subject had actually
completed the study. This type of analysis, often termed an “endpoint” analysis, introduces
its’ own set of problems in that (a) all subjects are treated equally regardless of the actual
intensity of their treatment over the course of the study, and (b) the actual responses that
would have been observed at the end of the study, if the subject had remained in the study
until its’ conclusion, may in fact be quite different from the response made at the time of
discontinuation. Returning to our example of the study in which subjects discontinue when
they feel that they have received full treatment benefit, an endpoint analysis might miss the
fact that some of these subjects may have had a relapse had they remained on treatment.
Many other objections have been raised about these two simple approaches of handing
missing data in longitudinal studies.
 PREFACE XV

In an attempt to provide a more general treatment of longitudinal data, with more

realistic assumptions regarding the longitudinal response process and associated missing
data mechanisms, statistical researchers have developed a wide variety of more rigorous
approaches to the analysis of longitudinal data. Among these the most widely used
includemixed-effects regression models [Lairdand Ware, 19821,and generalizedestimating
equation (GEE) models [Liang and Zeger, 1986; Zeger and Liang, 19861. Variations of
these models have been developed for both discrete and continuous outcomes and for
a variety of missing data mechanisms. The application of these models to the analysis
of clustered andor longitudinal data is the primary focus of this book. The primary
distinction between the two general approaches is that mixed-effects models are “full-
likelihood” methods and GEE models are “partial-likelihood” methods. The advantage of
statistical models based on partial-likelihoodis that (a) they are computationally easier than
full-likelihood methods, and (b) they generalize quite easily to a wide variety of outcome
measures with quite different distributional forms. The price of this flexibility, however, is
that partial likelihood methods are more restrictive in their assumptions regarding missing
data than their full-likelihood counterparts. In addition, full-likelihood methods provide
estimates of person-specific effects (e.g., person-specific trend lines) that are quite useful
in understanding inter-individual variability in the longitudinal response process and in
predicting future responses for a given subject or set of subjects from a particular subgroup
(e.g., a county, or a hospital, or a community). The distinctions between the various
alternative statistical models for analysis of longitudinal data will be fully explored in the
following chapters.
The outline of this book is as follows. Chapter 1 provides an overview of some of
the history of analysis of longitudinal data and brief discussion of the various methodolo-
gies that have been used in analysis of longitudinal data. Chapter 2 presents univariate
analysis of variance (ANOVA) models for analysis of repeated measurements. Chapter
3, presents multivariate analysis of variance (MANOVA) methods for analysis of repeated
measurements.
Chapter 4 provides the conceptual and statistical foundation for mixed-effects regression
models (MRM), which is the primary focus of this book. This chapter presents the general
random intercept and trend growth curve model with various extensions including group
structures and time-varying covariates. Chapter 5 considers curvilinear and polynomial
trend MRMs, including discussion and illustration of the use of orthogonal polynomials.
Chapter 6 presents covariance pattern models (CPMs) for analysis of longitudinal data.
These models can be viewed as extending the MANOVA approach, while allowing for
incomplete data across time and various forms for the error variance-covariance matrix
(i. e., autoregressive, Toeplitz, unstructured). Chapter 7 provides a further generalization
of the MRM to the case of autocorrelated residuals. This chapter also discusses selection
and comparison of the many possible models for the variance-covariance matrix of the
repeated measures.
Chapter 8 introduces generalized estimating equations (GEE) models for analysis of
longitudinal data. Since this class of models can be viewed as extending generalized linear
models (GLMs) to the case of correlated data, both continuous and categorical outcomes
are considered. Conceptual and statistical foundations of GEE models are presented, with
comparisons to MRMs and CPMs. Chapter 9 presents MRMs for anaylsis of dichotomous
longitudinal data. Using ordinary logistic regression as a starting point, the addition of
random effects to the model is described, as are the consequences this has on the scale and
interpretation of the regression coefficients in the model. Some discussion is paid to the
distinction between the population-averagedestimates of the GEE models and the subject-
xvi PREFACE

specific estimates of the categorical MRMs. Chapter 10 then extends the model for ordinal
outcomes and presents the many varieties of ordinal MRMs, including proportional and
nonproportional odds models. Chapter 11 presents MRMs for nominal responses, when
the categorized responses are unordered.
Chapter 12 describes models for the analysis of count data using a Poisson process. Such
data commonly occur in health services research where the number of service visits is an
outcome of interest. The models are then further generalized to the case of a zero-inflated
Poisson (ZIP) model, which segments the response process into (a) a logistic regression
model for the presence or absence of utilization and (b) a Poisson regression model for the
intensity of utilization, conditional on use.
Chapter 13 presents three-level generalization of the previously presented two-level
linear and nonlinear MRMs. An example is a multi-center longitudinal clinical trial in
which subjects are nested within centers and repeatedly measured over time. Finally,
Chapter 14 presents a detailed overview and discussion of the problem of missing data in
analysis of longitudinal data. We present an overview of the various approaches to this
problem, and we discuss in detail application of selection and pattern mixture models.
Throughout the book, applications of these methods for analysis of longitudinal data
are emphasized and extensively illustrated using real examples. However, we have chosen
not to focus on software in the book, though some syntax examples are provided. Many
programs are available for the analyses presented in this book including SAS, SPSS,
STATA, SYSTAT, HLM, MLwiN, MIXREGhfIXOR, and Mplus. To accompany this
book, we have decided to post several datasets and computer syntax examples on the
website h t t p : //www .u i c . edu/-hedeker/long .html. Our aim is to keep these syntax
examples current as new versions of the software programs emerge.
Most of the material from this book grew out of a class on Longitudinal Data Analysis,
taught at the University of Illinois at Chicago. This semester-long class typically covers
all of the material in Chapters 1 through 9 and 15. Overheads and additional materials
from this course are available at h t t p : //www . u i c . edu/classes/bstt/bstt513. The
students in this class are diverse, as is their level of statistical background. As a result,
this book does not assume a great deal of statistical knowledge. Essentially, one should
have a good knowledge of multiple regression and ANOVA modeling, along with some
knowledge of matrix algebra, maximum likelihood estimation, and logistic regression.
Several friends, colleagues, and students have helpedus with this book. In particular, we
thank R. Darrell Bock for teaching us everything that we know in statistics (though he is not
responsible for our errors of learning). Our colleague Hakan Demirtas provided extensive
help on the chapter on missing data. A big thanks goes to Ann Hohmann at N.I.M.H.,
who has been an incredible supporter of our work for many years. We are grateful for
the support provided by grants MH56146, MH65556, MH66302, and MH01254. Several
colleagues and students helped very much in reading over, discussing, and correcting drafts
of the chapters. In this regard, thanks go to Michael Berbaum, Richard Campbell, Mark
Grant, Zeynep Isgor, Andrew Leon, Julie Tamar Shecter, Matheos Yosef, and an anonymous
reviewer. Additionally, we thank Subhash Aryal for help in the Latex preparation of this
book. Finally, we thank Steve Quigley and Susanne Steitz of John Wiley & Sons for their
assistance throughout the entire process of writing and completing this book.

HEDEKER AND ROBERT D. GIBBONS

DONALD
Chicago, Illinois
January, 2006
ACKNOWLEDGMENTS

We have used several datasets from behavioral and medical studies in this text. We are
thankful to several people for their graciousness in sharing their data with us: John Davis,
Jan Fawcett, Brian Flay, Richard Hough, Michael Hurlburt, Robin Mermelstein, Niels
Reisby, and Nina Schooler.

D. H. andR. D. G.

xvii
This Page Intentionally Left Blank
ACRONYMS

AIC Akaike Information Criterion

ANOVA Analysis of Variance
ANCOVA Analysis of Covariance
BIC Bayesian Information Criterion
CPM Covariance Pattern Model
cs Compound Symmetry
DHHS Department of Health and Human Services
EB Empirical Bayes
GEE Generalized Estimating Equations
GLM Generalized Linear Model
HDRS Hamilton Depression Rating Scale
HLM Hierarchical Linear Model
ICC Intraclass Correlation
IRT Item Response Theory
IOM Institute of Medicine
LOCF Last Observation Carried Forward
MANOVA Multivariate Analysis of Variance
MAR Missing at Random
xix
XX Acronyms

MCAR Missing Completely at Random

MCMC Markov Chain Monte Carlo
ML Maximum Likelihood
MLE Maximum Likelihood Estimate
MNAR Missing Not at Random
MRM Mixed-effects Regression Model
NAS National Academy of Sciences
OPO Organ Procurement Organization
OPTN Organ Procurement Transplantation Network
REML Restricted Maximum Likelihood
SSCP Sum of Squares and Cross Products
SSRI Selective Serotonin Reuptake Inhibitor
TCA Tricyclic Antidepressant
UN Unstructured
ZAP Zero-altered Poisson
ZIP Zero-inflated Poisson
CHAPTER 1

INTRODUCTION

1.1 ADVANTAGES OF LONGITUDINAL STUDIES

There are numerous advantages of longitudinal studies over cross-sectional studies. First,
to the extent that repeated measurements from the same subject are not perfectly correlated,
longitudinal studies are more powerful than cross-sectional studies for a fixed number of
subjects. Stated in another way, to achieve a similar level of statistical power, fewer subjects
are required in a longitudinal study. The reason for this is that repeated observations from
the same subject, while correlated, are rarely, if ever, perfectly correlated. The net result is
that the repeatedmeasurements froma single subject provide moreindependentinformation
than a single measurement obtained from a single subject.
Second, in a longitudinal study, each subject can serve as hisher own control. For
example, in a crossover study, each subject can receive both experimental and control
conditions. In general, intra-subject variability is substantially less than inter-subject
variability, so a more sensitive or statistically powerful test is the result. As previously
mentioned, in naturalistic or observational studies, the primary intervention of interest
may also be time-varying, so that naturalistic intra-subject changes in the intervention
can be related to changes in the outcome of interest within individuals. Again, the net
result is an exclusion of between-subject variability from measurement error which results
in more efficient estimators of treatment-related effects when compared to corresponding
cross-sectional designs with the same number and pattern of observations.
Third, longitudinal studies allow an investigator to separate aging effects ( ie., changes
over time within individuals), from cohort effects ( i.e., differences between subjects at

LongitudinalDara Analysis. By Donald Hedeker and Robert D. Gibbons 1

Copyright 02006 John Wiley & Sons, Inc.
2 INTRODUCTION

baseline). Such cohort effects are often mistaken for changes occurring within individuals.
Without longitudinal data, one cannot differentiate these two competing alternatives.
Finally, longitudinal data can provide information about individual change, whereas
cross-sectional data cannot. Statistical estimates of individual trends can be used to better
understand heterogeneity in the population and the determinants of growth and change at
the level of the individual.

1.2 CHALLENGES OF LONGITUDINAL DATA ANALYSIS

Despite their advantages, longitudinal data are not without their challenges. Observations
are not, by definition, independent and we must account for the dependency in data using
more sophisticated statistical methods. The appropriate analytical methods are not as well
developed, especially for more sophisticated models that permit more general forms of
correlation among the repeated measurements. Often, there is a lack of available computer
software for application of these more complex statistical models, or the level of statistical
sophistication required of the user is beyond the typical level of the practitioner. In certain
cases, for example nonlinear models for binary, ordinal, or nominal endpoints, parameter
estimation can be computationally intensive due to the need for numerical or Monte Carlo
simulation methods to evaluate the likelihood of nonlinearmixed-effects regression models.
An added complication that arises in the context of analysis of longitudinal data is
the invariable presence of missing data. In some cases, a subject may be missing one of
several measurement occasions; however, it is more likely that there are missing data due
to attrition. Attrition, sometimes referred to as “drop-out,’’ refers to a subject removing
himself or herself from the study, prior to the end of the study. The data record for
this subject therefore prematurely terminates. Several simple approaches to this problem
have been proposed, none of which are statistically satisfactory. The simplest approach,
termed a “completer analysis,” limits the analysis to only those subjects that completed the
studymissing data,completer analysis. Unfortunately, the available sample at the end of
the study may have little resemblance to the sample initially randomized. Reasons for not
completing the study may be confounded with the effects that the study was designed to
investigate. For example, in a randomized clinical trial of a new drug versus placebo, only
those subjects that did well on the drug may complete the study, giving the potentially false
appearance of superiority of drug over placebo. The second simple approach is termed
“Last Observation Carried Forward” (LOCF) and involves imputing the last available
measurement to all subsequent measurement occasions. While things are somewhat better
in the case of LOCF versus completeranalyses, in an LOCF analysis, subjects treated in the
analysis as if they have had identical exposure to the drug may have quite different exposures
in reality or their experience on the drug may be complicated by other factors that led to
their withdrawal from the study that are ignored in the analysis. More rigorous statistical
alternatives based on mixed-effects regression models with ignorable and nonignorable
nonresponse are an important focus of this book. Nevertheless, the presence of missing
data, and its treatment in the statistical analysis, is a complicating feature of longitudinal
data, making analysis potentially far more complex than analysis of cross-sectional data.
The advantage, however, is that all available data from each subject can be used in the
analysis, leading to increased statistical power, the ability to estimate subject-specific
effects, and decreased bias due to arbitrary exclusion of subjects with incomplete response
or the simple imputation of values for the missing responses.
Random documents with unrelated
content Scribd suggests to you:
board what we had on shore, and made every thing ready for
weighing; accordingly, we began to unmoor in the afternoon.
Departure The 31st of December at four of the clock in the
from morning we weighed, and at six o’clock we left the
Bougainville bay, being towed by our boats. It was calm; at seven
Bay.
a light breeze sprung up at N. E. which became
more fresh in the day; the weather was clear till noon, when it
became foggy and rainy. At half an hour past eleven, being in the
middle of the strait[80], we discovered, and set the Cascade bearing
S. E. the Sugar Loaf S. E. by E. ½ E. Cape Forward[81] E. by N. Cape
Holland[82] W. N. W. ½ W. From noon till six in the evening we
doubled Cape Holland. It blew a light breeze, which Anchorage in
abating in the evening, and the sky being covered, I Fortescue Bay.
resolved to anchor in the road of Port Galant, where we anchored in
sixteen fathoms, coarse gravel, sand and small coral; Cape Galant
bearing S. W. 3° W[83]. We had soon reason to congratulate
ourselves on being in safety; for, during the night, it rained
continually, and blew hard at S. W.
1768. We began the year 1768 in this bay, called Bay
January. Fortescue, at the bottom of which is Port Galant[84].
The plan of the bay and port is very exact in M. de Gennes. We have
had too much leisure to confirm it, having been confined there for
three weeks together, by such weather as one cannot form any idea
of, from the worst winter at Paris. It is but just to let Account of the
the reader partake in some measure of the obstacles we
disagreeable circumstances on these unlucky days, met with.
by giving the sketch of our stay in this place.
Vestiges we My first care was to send out people to view the
found of the coast as far as Bay Elizabeth, and the isles with
passage of which the straits of Magalhaens are full in this part.
English ships.
From our anchoring-place we perceived two of
these isles, which Narborough[85] calls Charles and Monmouth.
Those which are farther off he calls the Royal Isles, and the
westermost of all, he names Rupert Island. The west winds
preventing us from making sail, we moored with a stream-anchor.
The rain did not keep our people from going on shore, where they
found vestiges of the passage and touching of English ships; viz.
some wood, lately sawed and cut down; some spice-laurel trees[86],
lately stripped of their bark; a label of wood, such as in marine
arsenals, are generally put upon pieces of cloth, &c. on which we
very distinctly read the words, Chatham, March, 1766; they likewise
found upon several trees, initial letters and names, with the date of
1767.
Astronomical M. Verron, who had got all his instruments carried
and nautical upon the peninsula that forms the harbour, made an
observations. observation there at noon, with a quadrant; and
found 53° 40′ 41″ S. lat. This observation, and the bearings of Cape
Holland, taken from hence; and those of the same cape, taken the
16th of December, upon the point from Cape Forward, determine the
distance of Port Galant to Cape Forward, to twelve leagues. Here he
likewise observed, by the azimuth-compass, the declination of the
needle 22° 30′ 32″ N. E. and its inclination from the elevation of the
pole 11° 11′. These are the only observations he was able to make,
during almost a whole month; the nights being as gloomy as the
days. On the third of January, there was a fine opportunity, of
determining the longitude of this bay; by means of an eclipse of the
moon, which began here at 10 hours, 30′ in the evening; but the rain,
which had been continual in the day-time, lasted likewise through the
whole night.
The 4th and 5th the weather was intolerable; we had rain, snow, a
sharp cold air, and a storm; it was such weather as the Psalmist
describes, saying, Nix, grando, glacies, spiritus procellarum. On the
third I had sent out a boat on purpose, to endeavour to find out an
anchorage on the coast of Terra del Fuego; and they found a very
good one S. W. of the isles Charles and Monmouth. I likewise gave
them orders to observe the direction which the tide took in that
channel. With their assistance, and the knowledge of anchoring-
places, both to the northward and southward, I would have made
sail, even though the wind should be contrary; but it was never
moderate enough for me to do it. Upon the whole, during our stay in
this part of the straits, we observed constantly, that the tides set in as
in the part of the narrows or guts; i. e. that the flood sets to the
eastward, and the ebb to the westward.
Interview with On the 6th, in the afternoon, we had some fair
and description moments; and the wind too seemed to blow from S.
of the E. we had already unmoored; but the moment we
Pecherais.
were setting sail, the wind came back to W. N. W. in
squalls, which obliged us to moor again immediately. That day some
savages came to visit us. Four periaguas appeared in the morning,
at the point of Cape Galant; and, after stopping there for some time,
three advanced into the bottom of the bay, whilst one made towards
our frigate. After hesitating for about half an hour, they at last brought
her along-side of us, with repeated shouts of Pecherais. In this boat
were a man, a woman, and two children. The woman remained to
take care of the periagua; and the man alone came on board, with
much confidence, and with an air of gaiety. Two other periaguas
followed the example of the first; and the men came on board the
frigate with their children. Here they were soon very happy and
content. We made them sing, and dance, let them hear music; and,
above all, gave them to eat, which they did with much appetite. They
found every thing good; whether bread, salt meat, or fat, they
devoured what was offered to them. We found it rather difficult to get
rid of these troublesome and disgusting guests; and we could not
determine them to return to their periaguas, till we sent pieces of salt
flesh down into them, before their faces. They shewed no surprise,
neither at the sight of the ships, nor at the appearance of various
objects, that offered themselves to their eyes; this certainly shews,
that in order to be capable of being surprised at the work of art, one
must have some fundamental ideas of it. These unpolished men,
considered the master-pieces of human industry, in the same light as
the laws of nature and its phenomena. We saw them often on board,
and on shore, during several days which they stayed in Port Galant.
These savages are short, ugly, meagre, and have an
insupportable stench about them. They are almost naked; having no
other dress than wretched seal-skins, too little for them to wrap
themselves in; these skins serve them equally as roofs to their huts,
and as sails to their periaguas. They have likewise some guanaco-
skins; but they are in small number. Their women are hideous, and
seemed little regarded by the men. They are obliged to steer their
periaguas, and to keep them in repair; often swimming to them,
notwithstanding the cold, through the sea-weeds, which serve as a
harbour to these periaguas, at a pretty distance from the shore, and
scooping out the water that may have got into them. On the shore
they gather wood and shells, without the men partaking in any thing
of their labour; nor are those women, who have children at their
breast, exempted from their task. They carry their children on their
backs, folded in the skins, which serve them as dresses.
Their periaguas are made of bark, ill connected with rushes, and
caulked with moss in the seams. In the middle of each is a little
hearth of sand, where they always keep up some fire. Their arms are
bows and arrows, made of the wood of a holly-leaved berberry-bush,
which is common in the straits; the bow-string is made of a gut, and
the arrows are armed with points of stone, cut with sufficient skill; but
these weapons are made use of, rather against game, than against
enemies; for they are as weak as the arms, which are destined to
manage them. We likewise saw amongst them, some bones of fish,
about a foot long, sharp at the end, and toothed along one side. This
is, perhaps, a dagger; or rather, as I think, an instrument for fishing:
they fix it to a long pole, and use it as a harpoon. These Indians,
men, women, and children, live promiscuously in their huts, in the
middle of which they light a fire. They live chiefly on shell-fish;
however, they have likewise dogs, and nooses, or springes, made of
whalebone. I have observed, that they had all of them bad teeth;
and, I believe, we must attribute that to their custom of eating the
shell-fish boiling hot, though half raw.
Upon the whole, they seem to be good people; but they are so
weak, that one is almost tempted to think the worse of them on that
account. We thought we observed that they were superstitious and
believed in evil genii; and, among them, the same persons, who
conciliate the influence of those spirits, are their physicians and
priests. Of all the savages I ever saw, the Pecherais are those who
are most deprived of every convenience; they are exactly, in what
may be called, a state of nature; and, indeed, if any pity is due to the
fate of a man, who is his own master, has no duties or business to
attend, is content with what he has, because he knows no better, I
should pity these men; who, besides being deprived of what renders
life convenient, must suffer the extreme roughness of the most
dreadful climate in the world. These Pecherais, likewise, are the
least numerous society of men I have met with in any part of the
world; however, as will appear in the sequel, there are quacks
among them: but as soon as more than one family is together, (by
family, I understand father, mother, and children) their interests
become complicated, and the individuals want to govern, either by
force or by imposture. The name of family then changes into that of
society; and though it were established amidst the woods, and
composed only of cousins-german, a skilful observer would there
discover the origin of all the vices, to which men, collected into whole
nations, have, by growing more civilized, given names; vices that
caused the origin, progress, and ruin of the greatest empires. Hence
it follows, by the same principle, that in civilized societies, some
virtues spring up, of which those who border on a state of nature are
not susceptible.
The 7th and 8th the weather was so bad, that we could not by any
means go from on board; in the night we drove, and were obliged to
let go our sheet-anchor. At some intervals the snow lay four inches
deep on the deck; and, at day-break, we saw that all the ground was
covered with it, except the flat lands, the wetness of which melted
the snow. The thermometer was about 5° and 4°; but fell to two
degrees below the freezing-point. The weather was bad on the ninth
in the afternoon. The Pecherais set out in order to come on board
us. They had even spent much time at their toilet; I mean, they had
painted their bodies all over, with red and white spots: but seeing our
boats go from the ships, towards their huts, they followed them; but
one periagua came on board the Etoile. She stayed but a short time
there, and joined the others; who were very much the friends of our
people. The women were, however, all retired into one hut; and the
savages seemed uneasy, whenever one of our men attempted to go
in. They invited them rather to come into the other huts, where they
presented our gentlemen with muscles, which they sucked before
they gave them away. They got some little presents, which they
gladly accepted. They sung, danced, and appeared more gay, than
one might expect from savages, whose outward behaviour is
commonly serious.
Unlucky Their joy was but of very short duration. One of
accident, their children, about twelve years old, the only one
which befalls in the whole troop whose figure engaged our
one of them.
attention, was all at once seized with spitting of
blood, and violent convulsions. The poor creature had been on board
the Etoile, where the people had given him bits of glass not
foreseeing the unhappy effect, which this present might have. These
savages have a custom of putting pieces of talc into their throat and
nostrils. Perhaps their superstition combines some powers with this
kind of talisman; or, perhaps, they look on it as a preservative
against some sickness they are subject to. The child, probably, had
made the same use of this glass. His lips, gums, and palate, were
cut in several places, and he bled continually.
This accident spread consternation and mistrust amongst them.
They certainly suspected us of some bad action; for the first thing
their juggler did, was to strip the child immediately of a linen jacket,
which had been given him. He wanted to return it to the French; and
upon their refusing it, he threw it at their feet. However, another
savage, who, doubtless, loved clothes more than he feared
enchantments, took it up immediately.
The juggler first laid the child down upon his back, in one of the
huts; and, kneeling down between his legs, he bent himself upon
him, and with his head and hands pressed the child’s belly as much
as he could, crying out continually, without our being able to
distinguish any articulate sounds in his cries. From time to time he
got up, and seeming to hold the disease in his joined hands, he
opened them all at once into the air, blowing as if he wanted to drive
away some evil spirit. During this ceremony, an old woman in tears,
howled in the sick child’s ears, enough to make him deaf. This poor
wretch seemed to suffer as much from the remedy, as from the hurt
he had received. The juggler gave him some respite, and went to
fetch his habit of ceremony; after which, having his hair powdered,
and his head adorned with two white wings, like those on Mercury’s
cap, he began his rites again, with more confidence, but with no
better success. The child then appearing to be worse, our chaplain
administered baptism to him by stealth.
 The officers returned on board, and told me what had happened
on shore. I went thither immediately with M. de la Porte, our surgeon,
who brought some milk and gruel with him. When we arrived, the
patient was out of the hut; the juggler, who had now got a companion
in the same dress, had begun again with his his operation on the
belly, thighs, and back of the child. It was a pity to see them torment
the poor creature, who suffered without complaining. His body was
already bruised all over; and the doctors still continued to apply their
barbarous remedy, with abundance of conjurations. The grief of the
parents, their tears, the part which the whole troop took in this
accident, and which broke out in the most expressive signs, afforded
us a most affecting scene. The savages certainly perceived that we
partook of their distress; at least they seemed to be less mistrustful.
They suffered us to come near the patient; and our surgeon
examined his bloody mouth, which his father and another Pecherais
sucked alternately. We had much trouble to persuade them to use
milk; we were obliged to taste it before them several times; and,
notwithstanding the invincible objection of their jugglers, the father at
last resolved to let his son drink it; he even accepted a pot-full of
gruel. The jugglers were jealous of our surgeon; whom, however,
they seemed at last to acknowledge as an able juggler. They even
opened for him a leather bag, which they always wear hanging by
their side; and which contains their feathered cap, some white
powder, some talc, and other instruments of their art; but he had
hardly looked into it, when they shut it again. We likewise observed,
that whilst one of the jugglers was conjuring the distemper of the
patient, the other seemed to be busied solely in preventing, by his
enchantments, the effect of the bad luck, which they suspected we
had brought upon them.
We returned on board, towards night, and the child seemed to
suffer less; however, he was plagued with almost continual puking,
which gave us room to fear that some glass was got down into his
stomach. We had afterwards sufficient reason to believe our
conjectures had been true; for about two o’clock in the morning, we
on board heard repeated howls; and, at break of day, though the
weather was very dreadful, the savages went off. They, doubtless,
fled from a place defiled by death, and by unlucky strangers, who
they thought were come merely to destroy them. They were not able
to double the westermost point of the bay: in a more moderate
interval they set sail again; a violent squall carried them out into the
offing, and dispersed their feeble vessels. How desirous they were of
getting away from us! They left one of their periaguas, which wanted
a repair on the shore, Satis est gentem effugisse nefandam. They
are gone away, considering us as mischievous beings: but who
would not pardon their resentment on this occasion? and, indeed,
how great is the loss of a youth, who has escaped from all the
dangers of childhood, to a body of men so very inconsiderable in
number!
Continuation of The wind blew east with great violence, and
bad weather. almost without intermission, till the 13th, when the
weather was mild enough in day-time; and we had even conceived
hopes of weighing in the afternoon. The night between the 13th and
14th was calm. At half an hour past two in the morning we had
unmoored, and hove a-peak. At six o’clock we were obliged to moor
again, and the day was dreadful. The 15th, the sun shone almost the
whole day; but the wind was too strong for us to leave the harbour.
Danger which The 16th, in the morning, it was almost a calm;
the frigate is then came a breeze from the north, and we
exposed to. weighed, with the tide in our favour: it was then
ebbing, and set to the westward. The winds soon shifted to W. and
W. S. W. and we could never gain the Isle Rupert, with the
favourable tide. The frigate sailed very ill; drove to leeward beyond
measure; and the Etoile had an incredible advantage over us. We
plyed all day between Rupert island, and a head-land of the
continent, which we called the Point of the Passage, in order to wait
for the ebb; with which I hoped either to gain the anchoring-place of
Bay Dauphine, upon the isle of Louis le Grand, or that of Elizabeth
bay[87]. But as we lost ground by plying, I sent a boat to sound to the
S. E. of Rupert’s-island, intending to anchor there, till the tide
became favourable. They made signal of an anchoring-place, and
came to a grapnel there; but we were already too much fallen to
leeward of it. We made one board in-shore, to endeavour to gain it
on the other tack; the frigate missed stays twice; and it became
necessary to wear; but at the very moment when, by the
manœuvres, and by the help of our boats, she began to wear, the
force of the tide made her come to the wind again; a strong current
had already carried us within half a cable’s length of the shore. We
let go our anchor in eight fathom: the anchor, falling upon rocks,
came home, and our proximity to the shore did not allow us to veer
away cable. We had now no more than three fathom and a half of
water a-stern; and were only thrice the length of the ship from the
shore, when a little breeze sprung up from thence, we immediately
filled our sails, and the frigate fell to leeward: all our boats, and those
of the Etoile, which came to our assistance, were a-head, towing her.
We veered away our cable, upon which we had put a buoy; and near
half of it was out, when it got foul between decks, and stopt the
frigate, which then ran the greatest danger. We cut the cable, and by
the prompt execution of this manœuvre, we saved the ship. The
breeze at length freshened; and, after having made two or three
unprofitable boards, I returned to Port Galant, where we anchored
again in twenty fathom oozy bottom. Our boats, which I left to weigh
our anchor, returned towards night with it and the cable. Thus this
appearance of fine weather served only to give us cruel alarms.
Violent The day following was more stormy than all the
hurricane. preceding ones. The wind raised a mountainous sea
in the channel; and we often saw several waves run in contrary
directions. The storm appeared to abate towards ten o’clock; but at
noon a clap of thunder, the only one we ever heard in this strait, was
as it were the signal at which the wind again began to blow with
more violence than in the morning. We dragged our anchor, and
were obliged to let go our sheet-anchor, and strike our lower-yards
and top-masts. Notwithstanding this, the shrubs and plants were now
in flower, and the trees afforded a very brilliant verdure, which
however was not sufficient to dispel that sadness which the repeated
sight of this unlucky spot had cast over us. The most lively temper
would be overcome in this dreadful climate, which is shunned by
animals of every element, and where a handful of people lead a
languid life, after having been rendered still more unfortunate by their
intercourse with us.
Assertion On the 18th and 19th there were some intervals
concerning the between the bad weather: we weighed our sheet
channel of anchor, squared our yards, and set up our top-
Sainte Barbe masts; and I sent the Etoile’s barge, which was in so
discussed.
good a condition as to be able to go out in almost
any weather, to view the channel of Sainte Barbe. According to the
extract M. Frezier gives of the Journal of M. Marcant, who
discovered and passed through it, this channel must bear S. W. and
S. W. by S. from Bay Elizabeth. The barge returned on the 20th, and
M. Landais, who commanded it, informed me, that having followed
the track and marks taken notice of by M. Marcant, he had not found
the true mouth, but only a narrow channel, closed by shoals of ice
and the land, which it is the more dangerous to follow, as it has not a
single good anchoring place, and as it is crossed in the middle by a
sand covered with muscles. He then went all round the isle of Louis
le Grand to the southward, and re-entered the channel of
Magalhaens, without having found any other. He only saw a fine bay
on the coast of Terra del Fuego, which is certainly the same with that
which Beauchesne calls Nativity Bay. Upon the whole, by going S.
W. and S. W. by S. from Bay Elizabeth, as Frezier says that Marcant
did, you must cut through the middle of the isle of Louis le Grand.
This information gave me room to believe that the It is inhabited,
channel of Sainte Barbe was opposite the very bay notwithstanding
where we now lay. From the top of the mountains its small size.
which surround Port Galant, we had often discovered southward of
the isles Charles and Monmouth, a vast channel, full of little islands,
and terminated by no land to the southward; but, as at the same time
we perceived another inlet southward of the isle of Louis le Grand,
we took that for the channel of St. Barbe, as being more conformable
to Marcant’s account. As soon as we were sure that this inlet was no
more than a deep bay, we no longer doubted that the channel of
Sainte Barbe was opposite Port Galant, southward of Charles and
Monmouth Islands. Indeed, reading over again the passage in
Frezier, and comparing it with his chart of the strait, we saw that
Frezier, according to Marcant’s report, places Elizabeth Bay, from
whence the latter set sail, in order to enter into his channel, about
ten or twelve leagues from Cape Forward. Marcant therefore must
have mistaken Bay De Cordes for Bay Elizabeth, the former lying
actually eleven leagues from Cape Forward, being a league
eastward of Port Galant: setting sail from this bay, and standing S. E.
and S. E. by S. he came along the westermost point of Charles and
Monmouth isles, the whole of which he took for the isle of Louis le
Grand; an error into which every good navigator may easily fall,
unless he is well provided with good directions: and then he stood
into the channel full of isles, of which we had a prospect from the top
of the mountains.
Utility which The perfect knowledge of the channel of Sainte
would accrue Barbe would be so much the more interesting, as it
from the would considerably shorten the passing of the straits
knowledge of
the channel of
of Magalhaens. It does not take much time to come
Sainte Barbe. to Port Galant; the greatest difficulty before you
come there, being to double Cape Forward, which is
now rendered pretty easy, by the discovery of three ports upon Terra
del Fuego: when you are once got to Port Galant, should the winds
prevent your taking the ordinary channel, if they be ever so little
upon the northerly points, the channel is open to you, opposite to this
port; in twenty-four hours you can then be in the South Seas. I
intended to have sent two barges into this channel which I firmly
believe to be that of Sainte Barbe; they would have completely
solved this problem, but the bad weather prevented their going out.
Exceeding The 21st, 22d, and 23d, squalls, snow, and rain,
violent squall. were continual. In the night between the 21st and
22d, there was a calm interval; it seemed that the wind afforded us
that momentary repose, only in order to fall harder upon us
afterwards. A dreadful hurricane came suddenly from S. S. W. and
blew with such fury as to astonish the oldest seamen. Both our ships
had their anchors come home, and were obliged to let go their sheet-
anchor, lower the lower yards, and hand the top-masts: our mizen
was carried away in the brails. Happily this hurricane did not last
long. On the 24th the storm abated, we got calm weather and sun-
shine, and put ourselves in a condition to proceed. Since our re-
entering Port Galant, we took several ton weight of ballast, and
altered our stowage, endeavouring by this means to make the frigate
sail well again; and we succeeded in part. Upon the whole,
whenever it is necessary to navigate in the midst of currents, it will
always be found very difficult to manage such long vessels as our
frigates generally are.
We leave Bay On the 25th, at one o’clock in the morning, we
Fortescue. unmoored, and hove a peek; at three o’clock we
weighed, and were towed by our boats; the breeze was northerly; at
half past five it settled in the east, and we set all our top-gallant and
studding-sails, which are very seldom made use of here. We kept
the middle of the strait, following its windings, for which Narborough
justly calls it Crooked Reach. Between the Royal Isles and the
continent, the strait is about two leagues wide; the channel between
Rupert Isle and Point Passage, is not above a league broad; then
there is the breadth of a league and a half between the isle of Louis
le Grand and Bay Elizabeth, on the easterly point of which, there is a
ledge covered with sea weeds, extending a quarter of a league into
the sea.
Description of From Bay Elizabeth the coast runs W. N. W. for
the strait from about two leagues, till you come to the river which
Cape Galant to Narborough calls Bachelor, and Beauchesne, du
the open sea.
Massacre; at the mouth of which, is an anchoring-
place. This river is easily known; it comes from a deep valley; on the
west, it has a high mountain; its westerly point is low, wooded, and
the coast sandy. From the river Bachelor, to the entrance of the false
strait or St. Jerom’s channel, I reckon three leagues, and the bearing
is N. W. by W. The entrance of this channel seems to be half a
league broad, and in the bottom of it, the lands are seen closing in to
the northward. When you are opposite the river du Massacre, or
Bachelor, you can only see this false strait, and it is very easy to take
it for the true one, which happened even to us, because the coast
then runs W. by S. and W. S. W. till Cape Quade, which stretching
very far, seems to close in with the westerly point of the isle of Louis
le Grand, and leave no outlet. Upon the whole, the safest way not to
miss the true channel, is to keep the coast of Louis le Grand island
on board, which may be done without any danger. The distance of
St. Jerom’s channel to Cape Quade, is about four leagues, and this
cape bears E. 9° N. and W. 9° S. with the westerly point of the isle of
Louis le Grand.
 That island is about four leagues long, its north side runs W. N. W.
as far as Bay Dauphine, the depth of which, is about two miles, and
the breadth at the entrance, half a league; it then runs W. to its most
westerly extremity, called Cape St. Louis. As, after finding out our
error concerning the false strait, we run within a mile of the shore of
Louis le Grand island, we distinctly saw Port Phelippeaux, which
appeared to be a very convenient and well situated creek. At noon
Cape Quade bore W. 13° S. two leagues distant, and Cape St.
Louis, E. by N. about two leagues and a half off. The fair weather
continued all day, and we bore away with all our sails set.
From Cape Quade the strait runs W. N. W. and N. W. by W.
without any considerable turnings, from which it has got the name of
Long-Lane, or Long-Reach, (Longue Rue). The figure of Cape
Quade is remarkable. It consists of craggy rocks, of which, those
forming its highest summits, do not look unlike ancient ruins. As far
as this cape, the coasts are every where wooded, and the verdure of
the trees softens the aspect of the frozen tops of the mountains.
Having doubled Cape Quade, the nature of the country is quite
altered. The strait is inclosed on both sides by barren rocks, on
which there is no appearance of any soil. Their high summits are
always covered with snow, and the deep vallies are filled with
immense masses of ice, the colour of which bears the mark of
antiquity. Narborough, struck with this horrid aspect, called this part,
Desolation of the South, nor can any thing more dreadful be
imagined.
Being opposite Cape Quade, the coast of Terra del Fuego seems
terminated by an advanced cape, which is Cape Monday, and which
I reckon is about fifteen leagues from Cape Quade. On the coast of
the mainland, are three capes, to which we gave names. The first,
which from its figure, we called Cap Fendu, or Split Cape, is about
five leagues from Cape Quade, between two fine bays, in which the
anchorage is safe, and the bottom as good as the sheltered
situation. The other two capes received the names of our ships, Cap
de l’Etoile, three leagues west of Cap Fendu, and Cap de la
Boudeuse, in the same situation, and about the same distance from
the Cape of the Etoile. All these lands are high and steep; both
coasts appear clear, and seem to have good anchoring places, but
happily, the wind being fair for our course, did not give us time to
sound them. The strait in this part, called Longue Rue, is about two
leagues broad; it grows more narrow towards Cape Monday, where it
is not above four miles broad.
Dangerous At nine o’clock in the evening, we were about
night. three leagues E. by S. and E. S. E. off Cape
Monday. It always blew very fresh from east, and the weather being
fine, I resolved to continue my course during the night, making little
sail. We handed the studding sails, and close-reefed the top-sails.
Towards ten o’clock at night the weather became foggy, and the wind
encreased so much, that we were obliged to haul our boats on
board. It rained much, and the weather became so black at eleven,
that we lost all sight of land. About half an hour after, reckoning
myself a-breast of Cape Monday, I made signal to bring-to on the
star-board tack, and thus we passed the rest of the night, filling or
backing, according as we reckoned ourselves to be too near one or
the other shore. This night we have been in one of the most critical
situations during the whole voyage.
At half an hour past three, by the dawn of day, we had sight of the
land, and I gave orders to fill. We stood W. by N. till eight o’clock,
and from eight till noon, between W. by N. and W. N. W. The wind
was always east, a little breeze, and very misty. From time to time
we saw some parts of the coast, but often we entirely lost sight of it.
At last, at noon, we saw Cape Pillar, and the Evangelists. The latter
could only be seen from the mast-head. As we advanced towards
the side of Cape Pillar, we discovered, with joy, an immense horizon,
no longer bounded by lands, and a great sea from the west, which
announced a vast ocean to us. The wind did not continue E. it shifted
to W. S. W. and we ran N. W. till half an hour past two, when Cape
Victory bore N. W. and Cape Pillar, S. 3° W.
End of the After passing Cape Monday, the north coast
strait, and bends like a bow, and the strait opens to four, five,
description of and six leagues in breadth. I reckon about sixteen
that part.
leagues from Cape Monday to Cape Pillar, which
terminates the south coast of the straits. The direction of the channel
between these two capes, is W. by N. The southern coast is here
high and steep, the northern one is bordered with islands and rocks,
which make it dangerous to come near it: it is more prudent to keep
the south coast on board. I can say no more concerning these last
lands: I have hardly seen them, except at some short intervals, when
the fogs allowed our perceiving but small parts of them. The last land
you see upon the north coast, is Cape Victory (Cap des Victoires),
which seems to be of middling height, as is Cape Deseado (Desiré),
which is without the straits, upon Terra del Fuego, about two leagues
S. W. of Cape Pillar. The coast between these two capes is bounded
for near a league into the sea, by several little isles or breakers,
known by the name of the Twelve Apostles.
Cape Pillar is a very high land, or rather a great mass of rocks,
which terminates in two great cliffs, formed in the shape of towers,
inclining to N. W. and making the extremity of the cape. About six or
seven leagues N. W. of this cape, you see four little isles, called the
Evangelists; three of them are low, the fourth, which looks like a hay-
stack, is at some distance from the rest. They ly S. S. W. about four
or five leagues off Cape Victory. In order to come out of the strait, it
is indifferent whether you leave them to the south or northward; in
order to go in, I would advise that they should be left to the
northward. It is then likewise necessary to range along the southern
coast; the northern one is bordered with little isles, and seems cut by
large bays, which might occasion dangerous mistakes. From two
o’clock in the afternoon, the winds were variable, between W. S. W.
and W. N. W. and blew very fresh; we plyed till sun-setting, with all
our sails set, in order to double the Twelve Apostles. Departure
We were for a long while afraid we should not be taken from the
able to do it, but be forced to pass the night still in strait of
Magalhaens.
the straits, by which means we might have been
obliged to stay there more than one day. But about six o’clock in the
evening we gave over plying; at seven, Cape Pillar was doubled, and
at eight we were quite clear of the land, and advancing, all sails set,
and with a fine northerly wind, into the westerly ocean. We then laid
down the bearings whence I took my departure, in 52° 50′ S. lat. and
79° 9′ W. long, from Paris.
Thus, after constant bad and contrary weather at Port Galant, for
twenty-six days together, thirty-six hours of fair wind, such as we
never expected, were sufficient to bring us into the Pacific Ocean; an
example, which I believe is the only one, of a navigation without
anchoring from Port Galant to the open sea.
General I reckon the whole length of the strait, from Cape
observations Virgin (Mary) to Cape Pillar, at about one hundred
on this and fourteen leagues. We employed fifty-two days
navigation. to make them. I must repeat here, that from Cape
Virgin to Cape Noir, we have constantly found the flood tide to set to
the eastward, and the ebb to the westward, and that the tides are
very strong; that they are not by much so rapid from Cape Noir to
Port Galant, and that their direction is irregular there; that lastly, from
Port Galant to Cape Quade, the tides are violent; that we have not
found them very considerable from this cape to Cape Pillar, but that
in all this part from Port Galant, the water is subject to the same laws
which put them in motion from Cape Virgin; viz. that the flood runs
towards the easterly, and the ebb towards the westerly seas. I must
at the same time mention, that this assertion concerning the direction
of the tides in the strait of Magalhaens, is absolutely contrary to what
other navigators say they have observed there on this head.
However, it would not be well if every one gave another account.
Upon the whole, how often have we regretted that we had not got
the Journals of Narborough and Beauchesne, such as they came
from their own hands, and that we were obliged to consult disfigured
extracts of them: besides the affectation of the authors of such
extracts, of curtailing every thing which is useful merely in
navigation; likewise, when some details escape them that have a
relation to that science, their ignorance of the sea-phrases makes
them mistake necessary and usual expressions for vicious words,
and they replace them by absurdities. All their aim is to compile a
work agreeable to the effeminate people of both sexes, and their
labour ends in composing a book that tires every body’s patience,
and is useful to nobody[88].
Conclusions Notwithstanding the difficulties which we have met
drawn from with in our passage of the strait of Magalhaens, I
hence. would always advise to prefer this course to that of
doubling Cape Horn, from the month of September to the end of
March. During the other months of the year, when the nights are
sixteen, seventeen, and eighteen hours long, I would pass through
the open sea. The wind a-head, and a high sea, are not dangerous;
whereas, it is not safe to be under a necessity of sailing blindfold
between the shores. Certainly there will be some obstacles in
passing the straits, but this retardment is not entirely time lost. There
is water, wood, and shells in abundance, sometimes there are
likewise very good fish; and I make no doubt but the scurvy would
make more havock among a crew, who should come into the South
Seas by the way of Cape Horn, than among those who should enter
the same Seas through the straits of Magalhaens: when we left it, we
had no sick person on board.
END OF THE FIRST PART.
A

VOYAGE
ROUND THE

WORLD.
 PART the SECOND.

From our entrance into the Western Sea, to our return

to France.

Et nos jam tertia portat

Omnibus errantes terris et fluctibus æstas.
Virg. Lib. I.
 CHAP. I.
The run from the straits of Magalhaens to our arrival at the Isle of Taiti; discoveries
which precede it.
January. 1768. Direction of our From our entrance into the Western
track, after Sea, after some days of variable
leaving the winds, between S. W. and N. W. we
strait.
soon got S. and S. S. E. winds. I did
not expect to meet with them so soon; the west winds generally last
to about 30°; and I intended to go to the isle of Juan Fernandez, in
order to make good astronomical observations there. I intended by
this means to fix a sure point of departure, in order to cross this
immense ocean the extent of which is differently laid down, by
differently navigators. The early meeting with the S. and S. E. winds,
obliged me to lay aside this scheme of putting in there, which would
have prolonged my voyage.
Observation on During the first days, I stood as near west as
the position of possible; as well to keep my wind, as to get off from
the coasts of the coast; the bearings of which are not laid down
Chili.
with any certainty in the charts: however, as the
winds were then always in the western board, we should have fallen
in with the land, if the charts of Don George Juan, and Don Antonio
de Ulloa had been exact. These Spanish officers have corrected the
old maps of North America[89]; they make the coast run N. E. and S.
E. between Cape Corso and Chiloe; and that upon conjectures,
which they have certainly thought well-founded. This correction
happily deserves another; it was not a very comfortable one for
those navigators, who after coming out of the strait, endeavour to get
to the northward, with winds which constantly vary from S. W. to N.
W. by W. Sir John Narborough, after leaving the straits of
Magalhaens, in 1669, run along the coast of Chili, examining all the
inlets and creeks, as far as the river of Baldivia, into which he
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookluna.com