Lecture 1: The Knowledge Production Function Fabian Waldinger Waldinger () Knowledge Production 1 / 52
Practical Information Lecturer: Fabian Waldinger E-mail address: f.waldinger@warwick.ac.uk Offi ce hour: Thursdays: 4.30-5.30pm in S.2.92. Please come to the offi ce hour if you want to discuss questions on the course material. Most importantly: ask questions during the lectures starting today! Waldinger () Knowledge Production 2 / 52
Course Design We are going to discuss current research in economics of science. Papers discussed in the lecture cover areas in economics of innovation, labour economics, and economic history. New methodologies will be introduced before discussing a paper. We go very much in detail into one or two papers per week. Please read the papers (even if it is hard). Waldinger () Knowledge Production 3 / 52
Mini Course Content 1 The Knowledge Production Function 2 Peer Effects in Science 3 High-Skilled Migrants and Innovation Waldinger () Knowledge Production 4 / 52
The Knowledge Production Function: Lecture Content 1 Empirical methodology Differences-in-Differences. 2 Evidence on the knowledge production function from dismissals and bombings (Waldinger, 2014) Waldinger () Knowledge Production 5 / 52
Methodology: Differences-in-Differences We often want to evaluate the effect of a certain programme using pre and post-treatment data. Common problem: other factors (which affect treatment outcomes) also change from the pre to the post period (e.g. changes in the business cycle). Waldinger () Knowledge Production 6 / 52
Methodology: Differences-in-Differences Solution: find a control group that is unaffected by the treatment but otherwise behaves exactly the same. In that case we control for other changes between the pre-and the post period using the changes in the in the control group. Waldinger () Knowledge Production 7 / 52
Methodology: Differences-in-Differences The Differences-in-Differences Estimator Waldinger () Knowledge Production 8 / 52
Methodology: Difference-in-Differences Differences-in-Differences Estimator: Crucial Assumptions The key assumption is that treatment and control group would have the same time trend in the absence of the treatment. This does not mean that they have to have the same mean of the outcome! Diffi cult to verify but one usually uses pre-treatment data to show that the trends are the same. Waldinger () Knowledge Production 9 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression We can estimate the differences-in-differences estimator in a regression framework. Advantages: It is easy to calculate standard errors. We can control for other variables which may reduce the residual variance (reduces standard errors). It is easy to include multiple periods. We can study treatments with different treatment intensity. (e.g. varying increases in the marginal tax rate for different people). Simplest DiD regression model: Outcome it = β 1 + β 2 Treatment i + β 3 Post t + β 4 (Treatment Post) it + ε it Treatment: dummy variable = 1 if individual in treatment group. Post: dummy variable = 1 after treatment. Waldinger () Knowledge Production 10 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression Outcome it = β 1 + β 2 Treatment i + β 3 Post t + β 4 (Treatment Post) it + ε it β 4 is the differences-in-differences estimate. In control group: Pre-treatment: Outcome it = β 1 Post-treatment: Outcome it = β 1 + β 3 In treatment group: Pre-treatment: Outcome it = β 1 + β 2 Post-treatment: Outcome it = β 1 + β 2 + β 3 + β 4 Differences-in-Differences: [y 1T y 0T ] [y 1C y 0C ] = [(β 1 + β 2 + β 3 + β 4 ) (β 1 + β 2 )] [(β 1 + β 3 ) β 1 ] = β 4 Waldinger () Knowledge Production 11 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression Outcome it = β 1 +β 2 Treatment i +β 3 Post t +β 4 (Treatment Post) it +ε it Waldinger () Knowledge Production 12 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression Outcome it = β 1 +β 2 Treatment i +β 3 Post t +β 4 (Treatment Post) it +ε it Waldinger () Knowledge Production 13 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression Outcome it = β 1 +β 2 Treatment i +β 3 Post t +β 4 (Treatment Post) it +ε it Waldinger () Knowledge Production 14 / 52
Methodology: Difference-in-Differences Differences-in-Differences Regression Outcome it = β 1 +β 2 Treatment i +β 3 Post t +β 4 (Treatment Post) it +ε it Waldinger () Knowledge Production 15 / 52
The Production of Scientific Knowledge Empirical evidence on the production of scientific knowledge from my paper: "Bombs, Brains, and Science: The Role of Human and Physical Capital for the Creation of Scientific Knowledge". Many countries, especially China, invest heavily in their universities. What are the inputs that create successful research universities? Basic inputs in knowledge production are human capital (H) and physical capital (K), i.e. scientists and laboratories (Machlup, 1961) Q = f(h,k) Estimating production functions is challening: Inputs are chosen based on unobserved productivity shocks (Ackerberg et al. 2007) Selection: Star scientists may be attracted by highly productive departments. Measurement error Waldinger () Knowledge Production 16 / 52
The Paper s Contribution Use exogenous and temporary shocks to human and physical capital of German and Austrian science departments: 1 Human capital shock: Dismissal of scientists in Nazi Germany 2 Physical capital shock: Bombing of universities during WW II. Waldinger () Knowledge Production 17 / 52
Shock 1: Dismissal of Scientists in Nazi Germany Immediately after securing power in 1933 the Nazi Government dismissed all Jewish and politically unreliable persons from civil service. Waldinger () Knowledge Production 18 / 52
Shock 1: Dismissal of Scientists in Nazi Germany Immediately after securing power in 1933 the Nazi Government dismissed all Jewish and politically unreliable persons from civil service. "Law for the Restoration of the Professional Civil Service,7th of April 1933: 3: Civil servants who are not of Aryan descent are to be placed in retirement...(this) does not apply to offi cials who had already been in the service since the 1st of August, 1914, or who had fought in the World War... 4: Civil servants who, based on their previous political activities, cannot guarantee that they have always unreservedly supported the national state, can be dismissed from service. After the annexation of Austria in 1938 the dismissals were extended to Austrian universities. Waldinger () Knowledge Production 19 / 52
Shock 1: Examples of Dismissed Professors Waldinger () Knowledge Production 20 / 52
Shock 1: Examples of Dismissed Professors Waldinger () Knowledge Production 21 / 52
Shock 1: Examples of Dismissed Professors Waldinger () Knowledge Production 22 / 52
Shock 1: Dismissals Physics Waldinger () Knowledge Production 23 / 52
Shock 1: Dismissals Physics Waldinger () Knowledge Production 24 / 52
Shock 1: Dismissals Chemistry Waldinger () Knowledge Production 25 / 52
Shock 1: Dismissals Mathematics Waldinger () Knowledge Production 26 / 52
Shock 1: Dismissals in Different Universities Waldinger () Knowledge Production 27 / 52
Shock 1: Dismissals in Different Universities Waldinger () Knowledge Production 28 / 52
Shock 2: Bombing of German Universities Universities were never targeted by the Allied bombing campaign but suffered collateral damage. Destruction of German cities concentrated during the last years of the war. Monthly bombing in tons 0 50000 100000 150000 Bombs dropped over Germany 1939 1940 1941 1942 1943 1944 1945 Year Source: Webster and Frankland (1961), Annex Waldinger () Knowledge Production 29 / 52
Shock 2: Example University of Berlin Waldinger () Knowledge Production 30 / 52
Shock 2: Example TU Aachen Waldinger () Knowledge Production 31 / 52
Shock 2: Bombing of German Universities Waldinger () Knowledge Production 32 / 52
Shock 2: Bombing of German Universities Waldinger () Knowledge Production 33 / 52
Data I: Panel of Scientists 1926-1980 I collected data on all German and Austrian scientists from 7 volumes of Kürschners Deutscher Gelehrten Kalender for the years: 1926, 1931, 1940, 1950, 1961, 1970, 1980. Data include all university physicists, chemists, and mathematicians at 7 points in time. Data contain 5,716 scientists (2,456 chemists, 2,000 physicists, and 1,260 mathematicians). Waldinger () Knowledge Production 34 / 52
Data II: Output of Science Departments Merge publications and citations in (current and historical) top journals from ISI Web of Science to each scientist. Generate department level output measures for each department and year by summing individual output. 5 year window around the relevant date: for 1926 I use publications between 1923 and 1927. Normalize total department output to have 0 mean and variance 1 in each subject to ensure comparability across subjects. Waldinger () Knowledge Production 35 / 52
Data III: Dismissal Shock Collected data on dismissed professors from a number of sources: 1 List of Displaced German Scholars ; edited 1937 2 Secondary Sources: Biographisches Handbuch der deutschsprachigen Emigration nach 1933 Vol II: The arts, sciences, and literature other sources focusing on physicists, chemists, and mathematicians (very few additional names from those sources). Waldinger () Knowledge Production 36 / 52
Data III: List of Displaced German Scholars Waldinger () Knowledge Production 37 / 52
Data III: List of Displaced German Scholars Waldinger () Knowledge Production 38 / 52
Data IV: Bombing Shock No existing dataset includes university (or department) level destruction measures. Collect department level destruction measures from information in university archives. 1 Contact all university archivists to obtain information. 2 If not successful, personally consult university archives to find information on bombing destruction of science departments. Data measure the percentage of buildings used by scientists from a specific department that were destroyed during WWII. Waldinger () Knowledge Production 39 / 52
Shock 2: Example TU of Berlin Waldinger () Knowledge Production 40 / 52
Effects of Dismissals and Bombings are Estimated with a DiD Strategy Most basic differences in differences model to estimate effect of human capital shock would be: Output dt = β 1 + β 2 HCShock(1933-40) d *Post t + β 3 Department FE d + β 4 Post t + ε dt Here we have multiple time periods (7 overall) so we estimate a version of differences-in-differences that estimates a coeffi cient for each year: Output dt = β 1 + t β 2t HCShock(1933-40) d *Year t + β 3 Department FE d + β 4 Post t + ε dt Waldinger () Knowledge Production 41 / 52
Human Capital Shock Waldinger () Knowledge Production 42 / 52
Plotting Coeffi cients and 95% Confidence Intervals Persistence of Dismissal Shock Effect of Dismissing 10 %.5.4.3.2.1 0.1 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 43 / 52
Physical Capital Shock Waldinger () Knowledge Production 44 / 52
Plotting Coeffi cients and 95% Confidence Intervals Persistence of Bombing Shock Effect of 10 % Destruction.5.4.3.2.1 0.1 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 45 / 52
Both Shocks on the Same Scale Effect of Dismissing 10 %.5.4.3.2.1 0.1.2 Persistence of Dismissal Shock Effect of 10 % Destruction.5.4.3.2.1 0.1.2 Persistence of Bombing Shock 1930 1940 1950 1960 1970 1980 Year 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 46 / 52
Both Shocks on the Same Scale - Citation Weighted Publications Persistence of Dismissal Shock Persistence of Bombing Shock Effect of Dismissing 10 %.4.3.2.1 0.1.2 Effect of 10 % Destruction.4.3.2.1 0.1.2 1930 1940 1950 1960 1970 1980 Year 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 47 / 52
Losing High-Quality Scientists Waldinger () Knowledge Production 48 / 52
Plotting Coeffi cients and 95% Confidence Intervals Coefficient 1.2 1.8.6.4.2 0 1930 1940 1950 1960 1970 1980 Year all top 50th top 25th top 10th top 5th Waldinger () Knowledge Production 49 / 52
Understanding Persistence of the Human Capital Shock: Department Size Effect of Dismissing 1 Scientist 2 1.5 1.5 0.5 1 1.5 Persistence of Dismissal Shock Department Size 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 50 / 52
Understanding Persistence of the Human Capital Shock: Hiring Quality Coefficient.6.4.2 0 all top 50th top 25th top 10th top 5th 1930 1940 1950 1960 1970 1980 Year Waldinger () Knowledge Production 51 / 52
Understanding Persistence of the Human Capital Shock: Peer Effects Localized negative peer effects could have affected scientists in departments that lost scientists due to the dismissal. These scientists may in turn affect the productivity of new scientists who joined the department. We will look at peer effects in great detail during the next lecture. The short answer: peer effects not important in this context. Waldinger () Knowledge Production 52 / 52