3 ms·
I'm curious what caused the cost of the new electronic system to increase so much. I understand there could be a lot to it on the back end to ensure privacy an
by froindt 9y ago
I'm curious what caused the cost of the new electronic system to increase so much. I understand there could be a lot to it on the back end to ensure privacy and anonymity of the data collected, but it doesn't seem like it should be a huge deal technically. We're talking about ~325 million people and collecting demographic info and address info [1]. The IRS has far far more variables to collect info on, however they have fewer people. 325 million people is nothing compared to scaled companies like Facebook, Amazon, or Google (far more data points per person, far more people).
Any speculation on cause or other considerations I'm missing? Did a quick search on Google News and didn't find anything. All the companies I listed have huge teams, but I am still not seeing how the cost has exploded.
https://www.census.gov/history/pdf/2010questionnaire.pdf https://www.census.gov/history/pdf/2010questionnaire.pdf
- drc500free 9y agoPointing to the highest performing companies that have been selected by the market is not really a fair comparison for a government agency. It's more relevant to look at what a median tech company would end up doing, if they had to launch at scale without organic growth but had unlimited funding.
- jonnathanson 9y agoTotally fair, but to play devil's advocate for a split second, let's not forget that the Census Dept has to make people aware of itself and get people to give their info willingly and within timelines, given no real incentive to do so. Facebook, Amazon, and Google are collecting info that is willingly passed their way. We are probably talking about buying data entry and management bloatware from a big mediocre enterprise contractor and then integrating the monstrosity.
- drc500free 9y agoTotally agree... Private sector just isn't going to serve people with a negative ROI and stay in business. Facebook's costs would skyrocket if they had to chase down the last 10-20% of the population necessary for comprehensive coverage. These are all structural issues with solving the problem, before we even get to a procurement process that picks a mediocre LPTA contractor instead of Google.
- froindt 9y ago> Pointing to the highest performing companies that have been selected by the market is not really a fair comparison for a government agency. Fair point. All the companies I pointed out are flush with cash, making the annual budget of the census comically small. So if we want large publicly visible government software projects to start going well consistently, what would it take? Recent poorly run projects have included the failed launch of healthcare.gov which was "rescued" by SV techies and this project which is way over budget. There are many more which are smaller, less visible to the public, and I'm sure there are a set of projects which get done on time and on or below budget.
- Spooky23 9y agoThe government contracting system is fubar. Everything is done by contractors and has layers and layers of nonsense that has to be tracked and complied with. For a program like the census that includes everything from development to Helpdesk to field service nationwide. I bet the end user support alone is $30-50M through a web of contractors. Work like this also gets deliberately broken up in chunks to lock in the votes of influential congressmen. More chunks are less efficient and more expensive. When I worked for a non-federal US government, the process for a small business to get "on contract" to actually bid on many projects cost anywhere from $20-50k in legal, insurance and compliance costs.
- edmccard 9y ago>I'm curious what caused the cost of the new electronic system to nearly double Nitpick: the article mentions that the cost estimates "had ballooned by nearly 50%", which is not the same as doubling (which would have involved the estimates increasing by nearly 100 percent)
- froindt 9y agoThanks, fixed that. Got mixed up between reading the article and typing the comment.