Recommended
More Related Content
Similar to How does Age Influence IncomeECO490Danni SongNanc.docx
Similar to How does Age Influence IncomeECO490Danni SongNanc.docx (18)
More from wellesleyterresa
More from wellesleyterresa (20)
Recently uploaded
Recently uploaded (20)
How does Age Influence IncomeECO490Danni SongNanc.docx
- 1. How does Age Influence Income? ECO490 Danni Song Nancy Haskell () - How Does Age Influence Income? This is a more grammatically correct title. You should remove my name and remove the last line, which merely repeats the title. Introduction The aim of this study is to find out if an individual age has any influence on income How do age influence income? As workers gains more experience there should be a premium to compesate their skills Knowing the relationship between the age and income is important in determining salaries and also individuals can use it to project their future earnings In the current generation the young population has outnumbered old people Nancy Haskell () - You can remove the 2nd bullet point because it is already stated on the previous slide. Literature review According to Cloninger (2016), age is a natural cause of
- 2. income disparity and it can’t be easily affected by government policies. Bares (2016), increase in age leads to increase in income up to the age of 55 years and after this age the level of income begin to decline. Organizations reported that older workers are more expensive than younger ones. Parramore (2016), Human capital theory argues that experience can increase income Knowledge of how age affects income is useful in explaining income disparity among workers The past studies have not clearly explained the type of relationship between age and income. Age-earning relationship is important in explaining income levels in employment life-cycle Model Regression equation is represented below �= �0 +�1w+�2b+�3as+�4a1539+ �4a4069+� where; income =w, age=�0, black=�1w,Asian =�2b, age15_39= �3as, age40_69=�4a4069, and � =error component. Income is the independent variable, age is a dependent variable. Control variables are white, black, Asian, age15_39, age40_69 The prediction is that age affects income Nancy Haskell () - Try to make sure your text does not run into the purple area, where we cannot read it. You can remove the first bullet point. The regresison equation needs the numbers after the betas to be subscripts. Also, the last betas should be a 5 (you have 4 listed
- 3. twice). Income is your dependent variable. Age15-39 and Age40-69 are your key independent variables. Black, White, and Asian are your control variables. You should remove the part that says age = B0.... that is wrong. You want to predict how age affect income in the last bullet. Meaning, what is the sign on the coefficient estimates? Data The data was collected from US Bureau of Labor Statistics (2017). The data is a time series for a period of 15 years. Sample size is made up of 350 observations Data collected consist earnings from employees of different age. The data contain all the required variables Data Summary statistics of the studyData MeanStandard dev.Min25thpercentileMedian75thpercentileMaxIncome 52847.168646.4833679646693508365825274551Age1539.34.02. 291.328.336.345.396Age4069.37.02.279.363.375.384.43White.8 0.12.269.719.8175.888.972Black .11.09.006.035.083.161.38Asian .05.08.006.017.027.047.567Race .96.05.678.942.973.9911.026 Nancy Haskell () - You need to increase the size of the font in this table to be more readable.
- 4. Empirical results The fixed effect model has three equations The first equation uses age and constant as the only variables and the R2=0.994 The second regression equation use black, Asian and constant variables, the value of is lower at R2=0.991 The last equation tests all the variables and the R2=0.994 This shows that age affects income more than other variables in the model Nancy Haskell () - You need a table of regression results, instead of words on this slide. Empirical results From hypothesis testing, the Hausman test and the F-test both gives values of zero meaning that fixed effects should be included According to the OLS age affects income Control variables have little effect on income Robustness check shows that all variable results are within the range Variables age4069 and black have a high relationship with income Nancy Haskell () - Do not discuss the Hausman and F-test in the presentation. As in the prior slide, you need a table of regression results rather than words. Conclusion The main objective of the study was to find out if age affects income
- 5. This means that experienced workers are compensated more According to the results age and race (black) have a high relationship with the income. Asian variable and white have less relationship with income Nancy Haskell () - Try to add one more bullet point that offers some thoughts on future research, or ways to expand your study. Works cited Bares, A. (2016). Cafe Classic: The Age-Earnings Relationship Is Not What You Think. Compensation cafe, 2-3. Cloninger, D. O. (2016). What factors influence income inequality? Conversation, 2-3. Parramore, L. S. (2016). ECONOMY. 50 Is the New 65: Older Americans Are Getting Booted from Their Jobs and Denied New Opportunities:, 1-2. Statistics, U. B. (2017). Current Employment Statistics - CES (National). washinhton DC: U.S. Bureau of Labor Statistics. Model: Try to make sure your text does not run into the purple area, where we cannot read it. You can remove the first bullet point. The regresison equation needs the numbers after the betas to be subscripts. Also, the last betas should be a 5 (you have 4 listed twice).
- 6. Income is your dependent variable. Age15-39 and Age40-69 are your key independent variables. Black, White, and Asian are your control variables. You should remove the part that says age = B0.... that is wrong. You want to predict how age affect income in the last bullet. Meaning, what is the sign on the coefficient estimates? Empirical results: Do not discuss the Hausman and F-test in the presentation. As in the prior slide, you need a table of regression results rather than words. Conclusion: Try to add one more bullet point that offers some thoughts on future research, or ways to expand your study. ECO490 Danni Song Dr. Nancy How do Age Influence Income? Introduction As people age their level of income might be expected to increase or remain at the same level until retirement. As individual ages, their working energy decrease leading to a decrease in their level of income. However, they gain more job experience as compared with employees who have worked for few years. The relationship between age and income explains why experienced workers have difficult time trying to adjust to job loss because their earnings reflect special skills required in the industry. The aim of this study is to find out if there exist any premiums to compensate for the job experience acquired after many years of work. The population data will be collected
- 7. from a sample divided into age groups, the first group will comprise of workers aged between 15 to 39 years and the second group will be made up of individuals aged between 40 to 69. The data will be analyzed using a regression model to find out if there is any linear relationship between age and income. Finally the papers will use the statistical results to explain the relationship between the variables. Literature review Age-earning relationship is used to explain income growth in an employment life cycle. Many studies have been done in the past. According to Cloninger (2016), it is important to understand the factors that bring income inequality in our society. He argues that understanding the causes of income inequality can assist in reducing the gap. However, he states that age is a natural cause of income disparity and it can’t be easily affected by government policies. According to Bares (2016), increase in age leads to increase in income up to the age of 55 years and after this age the level of income begin to decline. Bares (2016) explained from a research carried out between 2000 and 2009 by collecting data from individuals from different age groups. As an individual goes up through the employment ladder they gain more experience and become more resourceful to the organization they are working with. A research conducted by Towers Perrin found out that as employees become old, the organization views them as more expensive and less productive as compared to young workers according to Bares (2016). The study explains only increase in salary due to works experience and also reduction in earnings for old people because they become less productive to the company. The aim of this study is to found out if there exists any linear relationship between the variables, an existing relationship between the variables is important in decision making. There are different explanations of the relationship between age and income; they include human capital theory, aging labor markets and organizational dynamics. Income rises as a worker
- 8. ages and goes up the ladder until mid-50s when he or she reaches the peak of his or her earning and then the level of earnings start to decrease. The decrease in income from mid 50s is attributed to decrease in productivity due to advancement in age. Parramore (2016) explains that workers over 50 years are perceived to have less energy and this lead to decrease in income as a worker gets closer to retirement. This shows that age affects individual income in the labor markets and there should be a special relationship between the variables. This study will expound on the past studies by developing a regression analysis to explain the relationship between age and income. Model A model is a simplified reality description; an economic model is designed to yield hypotheses about a testable economic behavior .A model represents an economic process by use of quantitative relationship between the variables (Howitt, 2014). In this study a regression model will be used to represent the relationship between the independent and the dependent variable. The regression model is made up of one dependent variable, one independent variable and four control variables. Control variables in the equation include white, black, Asian, age15_39, age 40_69. The econometric model will be represented as follows; (�= �0 +�1w+�2b+�3as+�4a1539+ �4a4069+�)(include equation on its own line), where income (w), income, age (�0), black(�1w), Asian (�2b), age15_39(�3as), age40_69(�4a4069), and � (error component). Individual’s age affects the level of income and the control variables will be monitored to protect the results from any internal interference and the equation will be computed using the ordinary least square method. Control variables are constants meaning that only age that will have impact on the income. (This is not correct; control variable also need to vary and be represented by data in the regression) This section also needs to discuss economic theory/ reasoning
- 9. behind the coefficient predictions in table 1. Variables Table 1A Dependent variable Income Key Independent Variables Age Other Control Variables White Black Asian Age15_39 Age40_69 Hypothesis testing Table1 Econometric model: �= �0 +�1w+�2b+�3as+�4a1539+ �4a4069+� Predictions: dy/dx d2y/dx2 d2y/dx2 Dependent Variable: Income (I) Key independent variables: Age(a) β1+ 2β2 Ē>0 β1>0
- 10. β2<0 n.a Other control variables: White(w) β3 n.a n.a Black (b) β4 n.a n.a Asian (as) β5 n.a n.a Age15_39(age1539) β6 n.a n.a Age40_69(age1539) β7 n.a n.a Data The data is from be collected from the US Bureau of Labour Statistics website. According to the US Bureau of Labour Statistics records, the data is collected over a long period of time and this provides detailed information about the income of individuals in dfferent age groups. Data fron the US Bureau of Labour Statistics is a time series covering many years, this will assist in making a more reliable conclusion. The data is
- 11. corresponding to all the theoritical variables without any biasness. There is information about all the variables including the control variables. For this study, data for 350 workers was collected from 2000 to 2015, the study will cover 15 years. Empirical results Table 2 Data Mean Standard dev Minimum 25th percentile Median 75th percentile Maximum Income 52847.16 8646.483 36796 46693 50836 58252 74551 Age1539 .34 .02 .291 .328 .336
- 15. 0.991 0.994 Need to explain what is different across all these regressions and appropriately label the top roll of the table. R squared measures how close variables are fitted in the regression equation; a high value of Squared shows a close relationship between the variables. The R squared of 0.994 shows that age affects individual income. Change in age will lead to a change in the level of income. This means that others factors held constant a worker age have an impact on his or her income due to their additional skills. The error component and the autocorrelations are low because there is only one independent variable. The results may be affected by heteroskedacity because as workers gets near their retirement age their level of income will reduce. But the results cannot be concluded until hypothesis test is carried out. Need to interpret the magnitude, economic & statistical significance of your regression result. Hypothesis testing Table 4 Hausman test Ho: �re= �fe x2 stat= 47.25 p value=0.0000 Conclusion Reject ho, therefore include fixed effects F-test Ho: all state fixed effects statistically insignificant F-state=110.19 P value=0.0000 Conclusion Reject ho, therefore include fixed effects After testing for p values the null hypothesis is rejected, meaning that fixed effects are needed in the regressions.
- 16. Robustness section Robustness check Table 5 VARIABLES FE FE FE age1539 1.912 1.912 [1.184] [1.184] age4069 -2.274*** -2.274*** [0.755] [0.755] black -1.581** -1.26 -1.581** [0.747] [0.836] [0.747] Asian
- 17. 0.01 -0.093 0.01 [0.096] [0.123] [0.096] Constant 11.282*** 11.163*** 11.282*** [0.545] [0.035] [0.545] State FE Y Y Y Year FE Y Y Y Observations 350 350 350 R-squared 0.994 0.991 0.994 You then need to discuss how these results in table5 compare to the results in table3. Error component cannot be estimated except by carrying
- 18. out estimation and then making residuals. To estimate the error component we require non statistical information. There is need to incorporate panel data to get the clear relationship between the variables. Abstract The aim of this study is to establish if individual’s age has any impact on their level of income. The data was collected through observation from the US Bureau of Labour Statistics, analyzed and tested to determine the level of relationship. Information on the linear relationship between age and income is important to explain the premium paid to additional experience in labor market. Skilled workers have an advantage at work place as compared to younger workers. The results of the study did not provide a clear relationship and there is need to include panel data to assist in predicting variables that vary over time. The p values from f-test and Hausman test are zero meaning that further research should be carried out to determine other variable that may assist in getting the correct relationship. Reference Bares, A. (2016). Cafe Classic: The Age-Earnings Relationship Is Not What You Think. Compesation cafe, 2-3. Cloninger, D. O. (2016). What factors influence income inequality? Conversation, 2-3. Howitt, D. (2014). The sage Dictionary of Statistics. London: Sage. Parramore, L. S. (2016). ECONOMY. 50 Is the New 65: Older Americans Are Getting Booted from Their Jobs and Denied New Opportunities:, 1-2. Statistics, U. B. (2017). Current Employment Statistics - CES (National). washinhton DC: U.S. Bureau of Labor Statistics.
- 19. Eco 490, Haskell Spring 2017 Research Project Information Packet 1 A. General Tips for Getting Started ¾ Start with a broad topic, but work to refine your research question by (1) geographic location, (2) time period, (3) specific event, (4) specific policy change, (5) demographic group, or (6) some combination of these. ¾ As you begin researching a question, keep track of related questions that arise as you go. Eventually one of these questions will be your final research question, but seldom is it the one you started with. ¾ Look seriously at journal article sources on a given topic before you have a clear question. Research is an iterative process – a topic, some reading, a question, more reading, other questions, more reading, more questions, etc. ¾ Browse data sources to see what type of information is available. Sometimes
- 20. variables or trends in data can spark interesting questions. ¾ Keep the research project in the back of your mind… news articles, other classes, and casual conversations can spark great research questions. B. Journal Resources ¾ Google Scholar ¾ EconLit ¾ Economic encyclopedias x New Palgrave Dictionary of Economics x International Encyclopedia of the Social and Behavioral Sciences ¾ Handbook chapters (e.g., Handbook of Labor Economics) ¾ Published literature reviews (especially useful for finding other sources and identifying open questions in the literature) x Journal of Economic Literature x Journal of Economic Perspectives (very undergraduate accessible) Tips: x Read the abstract, introduction, conclusion, tables, and then the “meat” of an academic journal article – this saves time and increases comprehension relative to reading “front-to-back” as you would a novel.
- 21. x Look at the citations listed in a given article to find related, possibly more relevant articles. x In Google Scholar, check the “cited by” link to find newer, related articles that cited a given article https://scholar.google.com/ http://web.a.ebscohost.com/ehost/search/basic?sid=5324ee35- f4ec-4a50-b6a3- 8529c0c5fef3%40sessionmgr4005&vid=0&hid=4201 http://www.dictionaryofeconomics.com/dictionary http://www.sciencedirect.com/science/referenceworks/97800804 30768 https://www.elsevier.com/books/book-series/handbooks-in- economics https://www.jstor.org/journal/jeconlite https://www.jstor.org/journal/jeconpers Eco 490, Haskell Spring 2017 Research Project Information Packet 2 C. Some Publically Available Data Resources (this is my no means an exhaustive list and students should not feel constrained to the items here.) ¾ Macroeconomic Data x FRED (Federal Reserve Bank of St. Louis: Federal Reserve Economic Data)
- 22. ¾ International Data x Penn World Tables x National Trade Data Bank ¾ U.S. Microeconomic Data x CPS (Current Population Survey) x PSID (Panel Survey of Income Dynamics) x SIPP (Survey of Income and Program Participation) x HRS (Health and Retirement Study Panel Survey of Income Dynamics) ¾ U.S. Government Sources x Census Bureau x Bureau of Economic Analysis x Bureau of Labor Statistics x National Center for Health Statistics D. Final Paper Guidelines (See syllabus for directions on early stages such as the proposal!) I. Introduction (≈½ - 1 page) x Motivate the question… Why is your paper interesting/worth reading? x State the research question(s)… In general, what hypotheses are you testing? x Explain, in broad strokes, how you plan to answer the
- 23. question. x Briefly summarize your key findings and relate them to important policy issues and/or the broader literature. x Give a “roadmap” for the remainder of the paper II. Literature Review (≈1- 1.5 pages) x Discuss other studies on this topic, and relate each article to your analysis. x To the extent possible, focus on methodology, data, and results (not just results) x Note whether your study brings up new ideas or expands on old ones. x Refer to authors, not paper names (e.g., “Goldin and Katz (2000) argue that… “). The title of the paper does not need to appear anywhere except the works cited. https://research.stlouisfed.org/fred2/ http://www.rug.nl/research/ggdc/data/pwt/ http://www.trade.gov/mas/ian/tradestatistics/ http://www.census.gov/programs-surveys/cps.html http://simba.isr.umich.edu/data/data.aspx
- 24. http://www.census.gov/programs-surveys/sipp/data.html http://hrsonline.isr.umich.edu/index.php?p=data http://www.census.gov/data.html http://www.bea.gov/ http://www.bls.gov/ http://www.cdc.gov/nchs/ Eco 490, Haskell Spring 2017 Research Project Information Packet 3 III. Model (≈1.5 pages) x Verbally and mathematically describe and explain the theory you’re analyzing. Focus on the dependent variable and key independent variables. x Note the general variables in your model (e.g., Y = f(x1, x2, x3…) and briefly why they are needed and whether the affect the outcome positively or negatively. x For key independent variables, make predictions about the signs of marginal effects (consider second derivatives and cross-partial derivatives as needed). Where appropriate, make any significant predictions about magnitudes (e.g., elastic or inelastic). Justify the predications based on economic theory.
- 25. x Specify and justify the specific econometric model (regression equation). Given the theory above, discuss the appropriate functional form and methodology (linear, log-linear, OLS, fixed-effects, instrumental variables, etc.) x Express the theory in terms of testable hypotheses from the primary regression equation. Note any other relevant hypotheses (e.g., changes in the magnitude of coefficient estimates for a specific subsample relative the primary specification). x Note any restrictions to your analysis (e.g., simplifying assumptions imposed between the theory and the empirical model, or ideas that aren’t testable due to data constraints) IV. Data (≈½ - 1 page) x Name the data source(s) and give salient characteristics and background info. x Note whether the data are a cross-section, time series, or longitudinal. x Discuss whether the data are appropriate. (Do data correspond to theoretical
- 26. variables? Are the sources reliable and unbiased?) x Describe and justify any selection criteria used to narrow the sample. x Provide information on variables names, units of measurement, and key summary statistics. Note any anomalies or interesting features of the data. x Discuss potential problems that could affect the analysis (e.g., multicollinearity) Eco 490, Haskell Spring 2017 Research Project Information Packet 4 V. Empirical Results (≈1.5 pages) x Present and interpret your coefficient estimates. Discuss your results and compare them to your predicted hypotheses. Did results match predictions? x Address sign, magnitude (economic significance), and statistical significance. Focus primarily on your final regression model, although
- 27. address any secondary regression models as they relate to hypotheses presented in the model section. x Evaluate the explanatory power of your final model, including R2, adjusted R2, AIC, and any necessary considerations based on the error term analysis. (The error term analysis should consider issues such as normality, autocorrelation, heteroskedasticity, and the influence of outliers). x Discuss whether data limited your conclusions or ability to test hypotheses. VI. Robustness (≈½ page) x Present additional estimates to convince readers that your findings are “real.” x To the extent possible, address any concerns regarding omitted variables, alternative theories, biases in the data, sensitivity to outliers, endogeneity, etc. VII. Conclusion (≈½ - 1 page) x Briefly summarize your method and empirical results. Attempt to reconcile any differences between your predictions and the results.
- 28. x Put your findings in perspective relative to the literature. Attempt to reconcile any differences between your results and the literature. x Highlight the importance of your study. What does it add to existing knowledge? What important implications does it have for policy and/or for the literature? x Discuss how your research could be extended in the future. What is the next step in studying this theory? Abstract (≈ 100 words) x State your specific research question(s), and briefly explain your contribution to existing knowledge on the topic. x Summarize your method, data, and empirical results. References x Use any standard, accepted format for the works cited (e.g., APA). x Citations should include at least 5 peer reviewed journal articles.
- 29. Eco 490, Haskell Spring 2017 Research Project Information Packet 5 E. Table Guidelines (Example formatting) Table 1A: List of Variables Dependent Variable: Wages Key Independent Variables: Years of Experience Years of Schooling Other Control Variables: Gender Race (Black or White) Ethnicity (Hispanic) Dangerous Industry/Occupation Innate ability Notes: This is not a typical table to include in a paper, but will help facilitate model development after the initial proposal. A list of a variables might appear in an appendix with variable definitions. This is designed as an example, not necessarily a fully-specified regression. Table 1: Testable Alternative Hypotheses Econometric model: � = �0 + �1� + �2�2 + �3� + �4� + �5� + �6�� + �7� + �8� + � Predictions: ��
- 30. ��⁄ ��� ���⁄ ��� ����⁄ Dependent Variable: Wages (w) Key Independent Variables: Experience (E) �1 + 2�2�̅� > 0 �1 > 0 �2 < 0 n.a. Schooling (S) �3 > 0 �3 + �6� > 0 n.a. �6 < 0 Other Control Variables: Female (F) �4 > 0 n.a. n.a. Black (B) �5 < 0 n.a. �6 < 0 Hispanic (H) �7 < 0 n.a. n.a. Dangerous job (D) �8 > 0 n.a. n.a. Notes: For experience, the second derivative is actually 2�2 but we can ignore the constant 2 for the sake of predicting signs. In particular, note that “innate ability” appeared in Table 1A because we think it likely affects wages. However, “innate ability” does not appear in Table 1 because there is no appropriate variable to control for innate ability. If the data set included an appropriate proxy variable (e.g., IQ score) then we could include it. Or, if we had panel data, we could use individual
- 31. fixed effects to control for innate ability. (As you can see here, use table footnotes to clarify any necessary issues. Again, a table of this form is not used in published papers, but will facilitate early stages of research and model development.) Table 2: Sample Statistics Mean Standard Deviation Minimum 25th percentile Median 75th percentile Maximum Wages (w) Experience (E) Schooling (S) Female (F) Black (B) Hispanic (H) Dangerous job (D) Notes: Sample size includes 2,880 observations. (Use as needed for relevant information.) Eco 490, Haskell Spring 2017 Research Project Information Packet
- 32. 6 Table 3: Empirical Results Dependent variable: wages OLS Fixed Effects Experience 1.47*** (0.39) Experience squared Schooling Female Black Black*Schooling Hispanic
- 33. Dangerous job Constant Number of Observations 2,880 R2 Number of Individuals 0.46 1,440 Notes: Robust standard errors are reported in parentheses below each coefficient estimates. One, two, and three asterisks indicate statistical significance at the 10-, 5-, and 1-percent level, respectively. (Obviously the table would be appropriately filled in your version. The final regression results might include two, three, or four, main specifications – here I have shown one OLS specification and one Fixed Effects specification, where fixed effects are used to control for innate ability. Depending on the specifications, different information should appear in the bottom rows. For instance, if my specifications included dropping or adding independent variables, I should report R2 and adjusted- R2. If comparing models, I might consider including the AIC.) Table 4: Error Term Analysis
- 34. Initial Model Final Model Error Normality Autocorrelation Heteroskedasticity Key outliers or influential observations Notes: Details are omitted because this table will differ substantially by student and might only include results from specific statistical tests. Appropriate error term tests will vary depending on the regression and data set. The final model includes all variables and the appropriate functional form. The initial model might be a simpler functional form, have fewer variables, and/or have yet to correct for autocorrelation, etc. Eco 490, Haskell Spring 2017 Research Project Information Packet 7 Table 5: Robustness Dependent variable: wages Log-Linear Alt. Model 2 Alt. Model 3 Experience
- 35. Experience squared Schooling Female Black Black*Schooling Hispanic Dangerous job Constant Number of Observations R2 Notes: Robust standard errors are reported in parentheses below each
- 36. coefficient estimates. One, two, and three asterisks indicate statistical significance at the 10-, 5-, and 1-percent level, respectively. (Obviously the table would be appropriately filled in your version. Here, report regression results that can be compared to Table 3. For instance, one alternative model might consider log-linear instead of linear regressions in wages to see if results are robust. I might consider adding other explanatory variables or splitting the sample. Formatting should follow that of Table 3, but details will differ substantially by project.) F. A Couple of Reference Guides for Writing an Empirical Economics Research Paper Van Gaasbeck, Kristin A. 2007. Writing in Economics: Components of a Research Paper. Department of Economics, California State University, Sacramento, www.csus.edu/indiv/v/vangaasbeckk/resources/writing/comp.ht m. (Accessed 1/24/2016). Dudenhefer, Paul. 2014. A Guide to Writing in Economics. Department of Economics, Duke University. https://econ.duke.edu/uploads/media_items/a-guide- to-writing-in- economics.original.pdf. (Accessed 1/24/2016). x The PDF for this source is also on Isidore under our reading folder. I strongly suggest
- 37. reading Part II (all sections), Part III (all sections), and Part IV (sections 18-23). http://www.csus.edu/indiv/v/vangaasbeckk/resources/writing/co mp.htm https://econ.duke.edu/uploads/media_items/a-guide-to-writing- in-economics.original.pdf https://econ.duke.edu/uploads/media_items/a-guide-to-writing- in-economics.original.pdf