Applied Time Series Analysis for the Social Sciences (eBook)

Specification, Estimation, and Inference

Regina M. Baker (Autor)

eBook Download: EPUB

2025
342 Seiten
Wiley (Verlag)
978-1-119-01048-7 (ISBN)

Infos

EXPLORE THIS INDISPENSABLE AND COMPREHENSIVE GUIDE TO TIME SERIES ANALYSIS FOR STUDENTS AND PRACTITIONERS IN A WIDE VARIETY OF DISCIPLINES

Applied Time Series Analysis for the Social Sciences: Specification, Estimation, and Inference delivers an accessible guide to time series analysis that includes both theory and practice. The coverage spans developments from ARIMA intervention models and generalized least squares to the London School of Economics (LSE) approach and vector autoregression. Designed to break difficult concepts into manageable pieces while offering plenty of examples and exercises, the author demonstrates the use of lag operator algebra throughout to provide a better understanding of dynamic specification and the connections between model specifications that appear to be more different than they are.

The book is ideal for those with minimal mathematical experience, intended to follow a course in multiple regression, and includes exercises designed to build general skills such as mathematical expectation calculations to derive means and variances. Readers will also benefit from the inclusion of:

• A focus on social science applications and a mix of theory and detailed examples provided throughout
• An accompanying website with data sets and examples in Stata, SAS and R
• A simplified unit root testing strategy based on recent developments
• An examination of various uses and interpretations of lagged dependent variables and the common pitfalls students and researchers face in this area
• An introduction to LSE methodology such as the COMFAC critique, general-to-specific modeling, and the use of forecasting to evaluate and test models

Perfect for students and professional researchers in the political sciences, public policy, sociology, and economics, Applied Time Series Analysis for the Social Sciences: Specification, Estimation, and Inference will also earn a place in the libraries of post graduate students and researchers in public health, public administration and policy, and education.

REGINA M. BAKER has extensive experience teaching a time series course for the Inter-University Consortium for Political and Social Research summer program, the University of Oregon and the University of Notre Dame.

Textauszug

1
Introduction

1.1 Why Time Series and Why This Book?

Consider the following problems:

1. In testimony to US congressional committees in September of 2007, General David Petraeus noted that “the overall number of security incidents has declined in 8 of the past 12 weeks.”1 A graph of US troop fatalities in Iraq beginning in February of 2007, when the surge began, is consistent with General Petraeus’s claim and appears at first glance to provide dramatic support for the belief that the surge was effective; see the graph on the left‐hand side of Figure 1.1. The graph on the right‐hand side of the figure, though, raises questions about whether the declining number of deaths should be attributed to the surge. While it is true that there was a dramatic decrease in the number of US troop fatalities immediately after the beginning of the surge, there were equally dramatic decreases– and increases– in the number of troop fatalities at other times. Indeed, the series’ volatility is apparent, raising the possibility that the reduction attributed to the surge might just as reasonably be attributed to chance variation. In this case, the use of time series techniques is motivated by an important research design issue.
2. Studies of the relationship between politicians’ electoral success and the economic well‐being of their constituents are a staple in the voting behavior literature. Researchers attempting to extend these analyses by election results over time may find that the residuals from ordinary least squares (OLS) regression are autocorrelated. While introductory regression texts are likely to advise the use of generalized least squares (GLS) in this case, doing so ignores the possibility that interesting omitted variables are causing the autocorrelation. Reflexively applying GLS not only leaves that possibility unexplored, but will lead to biased estimates of the coefficients of interest if the omitted variables are also associated with the included variables.
3. Finally, researchers often find themselves with time series data relevant to their substantive interests, have some sense that they should use time series methods, but do not have the necessary training.

Figure 1.1 US troop fatalities in Iraq, 2003–2009

These common motivations for using time series techniques differ critically from the motivation assumed in most other time series presentations: the researchers did not begin with an attachment either to a particular time series technique, or to a theoretically‐derived equation representing the relationship between variables measured at different time periods. Instead, methods such as ARIMA2 modeling, generalized least squares regression, vector autoregression and autoregressive distributed lag models (including cointegration and error correction specifications) are presented separately, providing little information about the connections between them and the reasons for choosing one over another. Additionally, many econometricians begin with the assumption that their theory is correct – that they have included all relevant variables and controls – and seek only to estimate the model’s parameters.

Developments in time series analysis over the last few decades suggest that this process may be more of a hindrance to good time series work than a help. In the same way that researchers working with cross‐sectional data risk misinterpreting biased estimates if they do not consider possible confounding variables (omitted variables that are associated with both the dependent variable and one or more included explanatory variables), researchers working with time series data risk biased estimates if they do not utilize modeling techniques that give appropriate attention to delayed effects and complicated dynamic relationships. After some introductory remarks about time series data, a simple illustration will be provided in the next section.

1.2 Time Series: Preliminaries

1.2.1 What is a Time Series?

The critical feature of time series data is that observations have a logical order. For example, a subset of the Penn World Tables is presented in Table 1.1. Population, exports and imports are each a time series, and the accumulated data are logically ordered by the years for which they are reported. Ideally, time series observations are equally spaced – common examples include yearly data, monthly data, quarterly data, and daily data.

Table 1.1 An example of time series data.

Country

Year

Population

Exports(in millions)

Imports(in millions)

United States

1950

152 271 000

55 391

65 172

United States

1951

154 878 000

67 491

67 688

United States

1952

157 553 000

63 319

73 601

United States

1953

160 184 000

61 024

80 521

United States

1954

163 026 000

64 049

78 005

⋮

⋮

⋮

⋮

⋮

United States

2007

301 279 593

153 409

218 379

Exports and imports are shown in current international prices.

(Source: Heston et al., (2009)).

Notice that these data refer to a single unit (country, in this case) – the United States. Data collected over time for more than one country, as in Table 1.2, are referred to as “Pooled Cross‐section Time Series” data, and require attention to issues beyond those presented in this book.

Table 1.2 An example of pooled cross‐section time series data.

Country

Year

PopulationPopulation

Exports(in millions)

Imports(in millions)

Canada

1950

14 011 422

27 902

23 190

Canada

1951

14 330 675

30 497

26 708

Canada

1952

14 785 584

33 255

27 668

Canada

1953

15 183 375

32 930

30 067

Canada

1954

15 636 245

31 795

28 467

⋮

⋮

⋮

⋮

⋮

Canada

2007

32 935 961

52 895

51 945

Mexico

1950

28 485 180

43 307

74 556

Mexico

1951

29 296 235

40 583

93 704

Mexico

1952

30 144 317

42 059

91 260

Mexico

1953

31 031 279

48 927

87 186

Mexico

1954

31 959 113

51197

91 260

⋮

⋮

⋮

⋮

⋮

Mexico

2007

108 700 891

294 000

319 500

United States

1950

152 271 000

55 391

65 172

United States

1951

154 878 000

67...

Produkt-Details

• Erscheint lt. Verlag: 17.11.2025
• Reihe/Serie: Wiley Series in Probability and Statistics
• Sprache: englisch
• Themenwelt: Mathematik / Informatik ► Mathematik ► Statistik
• Themenwelt: Mathematik / Informatik ► Mathematik ► Wahrscheinlichkeit / Kombinatorik
• Schlagworte: ADL • Arch • ARIMA • Arima probability • error correction • GARCH • Gauss-Markov • GLS • lag operator algebra • ols • pooled cross-section time series model • social science time series • time series estimation • Time Series Forecasting • vector autoregression
• ISBN-10: 1-119-01048-9 / 1119010489
• ISBN-13: 978-1-119-01048-7 / 9781119010487