# The less obvious effect of hosting the Olympics on sporting performance

0 comment

### econometric model

We used the number of Olympic medals (total, male, female) for the country in the sport as an outcome variable. To measure the host effect for each Olympics separately, we created a dummy variable for each host country. For the average number of medals won by the control countries, we adopted the variable for the average number of medals won by the sample country at the Olympic Games. Average medal count is essential to get a reliable estimate for hosting dummies. Using a lagged dependent variable as a control for previous performance tends to introduce bias when fixed effects are employed in the model.27Since the number of events and the number of medals won varies, bidirectional, Olympic, and sport-level fixed effects were included in the regression model to address differences.12Due to the different specifications of the model, the model uses sport and game level fixed effects, as opposed to the country sport and game level fixed effects used by Singleton et al. Here are the baseline specs:

$${m}_{i,j,t}=\mathrm{\alpha}+{\upbeta}_{1}{d}_{i,t}+{\upbeta}_{2}{AM} _{i,j}+{\uptheta}_{t}+{\mathrm{\varphi}}_{j}+{\upvarepsilon}_{i,j,t},$$

(1)

where $${m}_{i,j,t}$$ country’s medal count I of j in sports t olympic games, $${d}_{i,t}$$ is the host country dummy variable I of t the Olympics and $${AM}_{i,j}$$is the average Olympic medal won in a sample of countries. I of j sports. $${\theta}_{t}$$ show the game, $${\varphi}_{j}$$ Sport-level fixed effects. $${\upvarepsilon}_{i,j,t}$$ is the disturbance term and is assumed to be heteroskedastic.

There are two factors that determine a country’s success at the Olympics: its economic resources and its talent pool.Four,Ten,11,16,17To control for these factors, we use logarithmic forms of GDP per capita and population size in our augmented model. Here are the extended specs:

$${m}_{i,j,t}=\mathrm{\alpha}+{\upbeta}_{1}{d}_{i,t}+{\upbeta}_{2}{AM} _{i,j}+{\upbeta}_{3}{\mathrm{ln}GDPpc}_{i,t}+{\upbeta}_{4}{\mathrm{ln}POP}_{i, t}+{\upbeta }_{5}{Communist\, bloc}_{i}+{\uptheta }_{t}+{\mathrm{\varphi }}_{j}+{\upvarepsilon }_{ i,j,t},$$

(2)

where $${\mathrm{ln}GDPpc}_{i,t}$$, $${lnPOP}_{i,t}$$ The main controlling variables are GDP per capita and country population size. I A year t. Use the logarithmic form of both variables. We also used dummies for the member states of the Soviet Union or countries under its influence. Previous studies have shown that these countries outperformed their socioeconomic counterparts in terms of Olympic medals even after the collapse of the Soviet Union.Four,Ten,17,twenty two. of $${communism\,block}_{i}$$ variable indicates the country I was included in this group.

To detect the pre- and post-effects of the Olympic host country, we extended models (1) and (2) with pre- and post-dummies as well as the host dummy. This model specification allowed us to distinguish the before and after effects from the medal surplus gained by hosting.

Most countries have never won an Olympic medal, so there are many zero observations in the sample (see Figure 2). 2), which can bias the estimation using OLS.Previous studies used the Tobit estimatorFour,Five,Ten,11,16,17,18 or zero-inflated beta regression15 Manage the zero observations problem. However, medal counts are only positive numbers exhibiting a Poisson or negative binomial distribution.28A zero-inflated Poisson (ZIP) or zero-inflated negative binomial (ZINB) model should be used to account for both problems.29.

A negative binomial overdispersion parameter α can be used to distinguish ZINB from ZIP models. For Poisson, the variance is equal to its mean. In contrast, negative binomial has overdispersion, so the variance is larger than the mean. Overdispersion is indicated when α > 0.Moreover, the larger α, the larger the negative binomial variance29We performed post-estimation tests on various estimators. First, a likelihood ratio test comparing ZINB and ZIP. Then a Vuong test comparing ZINB to the standard negative binomial model.

### data

We used data on the number of medals from the Summer Olympics from 1996 to 2021. Due to the change of government in the 1990s and the boycotts that preceded it, results from previous Olympics are not available for longitudinal analysis. Instead of aggregate data, we used sport-level medal counts to get more detailed information about each country’s Olympic performance.12,17The individual units are the countries associated with a particular sport (eg Afghanistan – Athletics) and the time dimension is the year of the Olympics in the dataset. We used data from countries with qualified athletes in the sport. In some countries he only participated in the Olympic Games once, so zero observations from all Olympic Games would bias the analysis.For socioeconomic indicators, we used data from the World Bank database30Since the Olympics are a four-year event, we calculate the four-year geometric mean of the year of the Olympics and the preceding three years to obtain more detailed information on the economic and social conditions of each country. Did. This method eliminates data volatility and bias due to erroneous data. The table shows summary statistics for the variables included in the analysis. A1 in the appendix.