In the early days of Pandemic Covid-19, the US government did not have vague instructions on the impact of the pandemic in the country, including how many people were hospitalized and how many died of this disease. This initial failure to track pandemics is probably the main reason why the pandemic becomes a national disaster.
Before the 2020 pandemic, the US has a pandemic preparation plan, This emphasizes the importance of decision-making driven by data. However, this plan is not the most important aspect of “decision making driven by data”: data. Without a clear plan of how to collect or analyze data, it is given that whatever plan will fail. There are many other serious problems, many of which are solved late to contain a pandemic.
Errors made along the way revealed some important lessons for all data scientists, not only those who work in epidemiology.
Lesson 1: Your boss may not understand data
One of the main reasons for many failed projects is the termination between scientists and data management. Business Professor Joshi and his colleagues handle this problem in MIT management articles why so many data sciences online training projects failed to send.
Another reason for the failed project: good data is inserted into a bad system. Many quality data is available at the beginning of a pandemic, but this data becomes broken along the way. The damage can be reduced if someone who is responsible is noting that there is a problem in the data path, but this does not happen. For example, the CDC published the dashboard in May 2020.
The dashboard contains significant errors, combining the results of a viral test (which is sized if someone has a virus) with an antibody test (which measures if someone has ever had a virus). CDC officials initially did not pay attention to the problem; Errors were immediately taken by news organizations, including Covid tracking projects, volunteer organizations launched by Atlantic. The organizational research also found incompatibility with the date of test and actual case, which ultimately caused the implementation of a haphazard pandemic policy.
Lesson 2: Models are difficult to apply to real life
In the fall of 2020, it was clear that the pandemic forecasting model failed miserably. The reason including a broad range of distance, including
High sensitivity estimates,
The assumption of incorrect modeling,
Not enough dimensions included in the model,
Bad data input,
It’s very easy for errors to seep into a model that considering a winding trip one piece of data Covid-19 must take: After the collection, the lab must first process the test results. No one knows the true accuracy of the Covid-19 test but this uncertain data continues “… through several layers of human observation, keyboard entries, and private computer systems,” before achieving “aging and inadequate data systems,” some. which even relies on fax documents.
Even if your model is built with anti-bullet accuracy, they don’t easily translate to real-life situations. “The difference between models and real life,” Country Carter Mechher, a medical advisor in the Department of Veteran Affairs, “… is that with the model, we can set the parameters as if they are known. In real life, this parameter is as clear as mud”.
Is there any solution?
If another hit pandemic, how do we ensure more accurate predictions? In addition to putting more checks in place to ensure data is not damaged along the road, one solution is to employ social scientists to work with data scientists. For business, decisions are driven by data usually rotate throughout the bottom line. The government has other factors that influence decisions, such as bureaucracy, allocation of resources, and politics. A data engineer can optimize for business income, but to take into account the other factors in the government, a social scientist is needed.
My mother knows the best: too many chefs damage drinks. Given that many small mistakes in Covid-19 data are compiled by a large number of people and the system, the result is a series of predictions that are no better than looking into 8 magic balls. Perhaps the real solution is to focus all state and federal testing systems, including the country-controlled laboratory.