Category Archives: Statistics

Confounding and the measurement of the MRT off peak initiative

The Singapore government announced a while back that they were going to start an initiative to try to reduce peak period crowds on our public rail system or MRT (Mass Rapid Transit). The initiative involved providing free and subsidised travel for passengers on selected trips during the morning off-peak period.

This initiative kicked off two days ago. Two days on, some people are wondering if it made any difference – trains seem as packed as they were before, and those who were already taking trains during the free travel periods have found little to no difference of the number of passengers from before.

But before we make any conclusions, aside from the fact that it’s only day two and it makes no sense to conclude any result, we have to realise that this initiative kicked in at a time filled with confounding variables.

What we’re trying to measure here is whether the government initiative has worked by reducing peak period travel. So we’re trying to see if there’s a relationship between the [Government Initiative] and [Fewer Peak Period Passengers]; or more precisely, whether [Government Initiative] caused [Fewer Peak Period Passengers].

A confounding variable is an additional variable (one we could and would rather do without) that obscures the relationship of the variables we’re trying to measure, because its introduction impacts the end result. Let me give you an example.

Let’s say there’s a group of people who are hard of hearing. You discover that they love listening to loud music and have, in fact, done so for at least the last five years. You might conclude that listening to loud music makes you hard of hearing.

But let’s say that you then discover that this group of people all used to operate jackhammers, and were subject to loud noises for most of their working lives. Would you be as confident of your conclusion now?

What if these people were in their 80s? Would that change your mind yet again?

Loud music, operating jackhammers, and age can all contribute to hearing loss. Drawing conclusions from this group to make predictions on hearing loss is going to be tough. You just can’t quite single out one cause for hearing loss.

So, as I was saying, any analysis of the MRT rides this week is definitely going to be badly confounded by (at least) the following:

  • People working from home due to the haze;
  • Parents bringing their children overseas as it’s the last week of the school holidays;
  • Children not taking the trains because it’s the school holidays, leading to;
  • People just trying out what it’s like to travel off peak, with the new initiative; and
  • People just trying out what it’s like to travel during peak periods, with the new initiative.

Confounding in business measurement

Confounding is a terrible thing to have when you’re trying to measure cause and effect. I remember having been involved in several performance measurement initiatives, all happening at the same time, designed to improve sales numbers.

The problem with such initiatives is that you could never really know how much of an impact a particular initiative had on the overall sales results. You could know the impact of all the initiatives put together, but any single one would probably have been affected by others because, as mentioned before, they were all happening at the same time.

It’s difficult to get management to agree putting off trying initiatives simply because you want to get a more accurate measurement. It’s like telling a get-rich-quick addict to try only one get-rich-quick scheme at a time to know what really works. It just doesn’t happen.

But when you don’t know what initiative works and what doesn’t, you can’t afford to drop even a single one of them. And juggling all of them can get pretty expensive.

The Reliability of Internet Marketing Research

I was doing some secondary research on the web to try to gather some statistics on small businesses and websites when I realised that there just wasn’t much reliable data around, and that the majority of the statistics on the web were referencing themselves (this is like when an article on website A would point to 50% of small businesses not having websites in 2011, a statistic it obtained from an article written in 2009 on website B, which got its information from website C, which was incidentally quoting an unconfirmed “Internet Research expert” who wrote it on some tech forum, citing some old and unconfirmed piece he remembered reading a couple of years ago).

Reminded me of an article I read on how Wikipedia was subject to these sorts of self-referencing too.