1 00:00:11,110 --> 00:00:17,650 In this lecture, we're going to discuss a contemporary application of the simple moving average, since 2 00:00:17,650 --> 00:00:22,930 this is still fresh in our minds, I think it's a great opportunity to see how basic concepts like these 3 00:00:23,140 --> 00:00:24,690 get applied in the real world. 4 00:00:25,420 --> 00:00:32,290 So covid-19, as you know, many data sources use a seven day rolling average to report the trend of 5 00:00:32,290 --> 00:00:33,340 new accounts. 6 00:00:33,970 --> 00:00:38,620 Using what you've learned in this section, you have a kind of insider view of the pros and cons of 7 00:00:38,620 --> 00:00:39,490 this approach. 8 00:00:40,630 --> 00:00:42,650 On one hand, this is very useful. 9 00:00:43,120 --> 00:00:46,360 We've now seen real world reasons for statistical noise. 10 00:00:46,720 --> 00:00:51,300 For example, sometimes there are data entry errors that needs to be corrected later. 11 00:00:52,180 --> 00:00:53,820 This will skew the case counts. 12 00:00:54,370 --> 00:01:00,100 Maybe they'll be undercounted one day and then overcounted another day to correct for the previous undercounting. 13 00:01:00,910 --> 00:01:06,490 Another example is where a specific locations don't always report their accounts promptly to whatever 14 00:01:06,490 --> 00:01:09,530 central authority is providing their accounts to the public. 15 00:01:10,300 --> 00:01:13,210 In this case, again, this will skew the case counts. 16 00:01:13,630 --> 00:01:18,910 The counts might be undercounted one day because some locations forgot to send their counts and then 17 00:01:18,910 --> 00:01:22,430 overcounted the next day when the backlogged counts are included. 18 00:01:23,020 --> 00:01:26,580 We've already seen how accounts from two different sources can disagree. 19 00:01:27,280 --> 00:01:30,190 You'd think a simple task like counting would be error free. 20 00:01:30,400 --> 00:01:32,770 But these are the complexities of the real world. 21 00:01:33,460 --> 00:01:38,930 We've also seen situations where incompetent politicians direct people to simply lie. 22 00:01:39,490 --> 00:01:41,710 In this case, again, accounts are inaccurate. 23 00:01:42,580 --> 00:01:47,500 Now, simple moving averages will not help you in this situation, but they do help you when the counts 24 00:01:47,500 --> 00:01:49,270 are simply delayed or backlogged. 25 00:01:50,930 --> 00:01:56,060 So using what you've learned in this course, what can we say about these seven day rolling averages? 26 00:01:56,630 --> 00:02:01,350 Well, we know that while it can smooth out statistical noise, it also has a lag. 27 00:02:01,790 --> 00:02:06,680 So if you see a very high seven day rolling average, this means that, yes, things are bad. 28 00:02:07,040 --> 00:02:09,220 But they were already bad a few days ago. 29 00:02:09,680 --> 00:02:11,480 In other words, it's old news. 30 00:02:11,990 --> 00:02:14,960 You might think an exponential moving average might help. 31 00:02:15,200 --> 00:02:18,950 But remember, this data needs to be communicated to the public. 32 00:02:19,310 --> 00:02:24,650 Most members of the public do not have the capacity to understand what an exponential moving average 33 00:02:24,800 --> 00:02:25,680 even is. 34 00:02:26,450 --> 00:02:32,030 Furthermore, because it involves a tunable parameter, specifically the decay rate, it's not clear 35 00:02:32,030 --> 00:02:37,520 how you could be consistent with everyone else in the world so that the counts are more easily comparable. 36 00:02:39,050 --> 00:02:44,420 So I hope this lecture helped you understand one real world application of the simple moving average 37 00:02:44,420 --> 00:02:50,540 as it pertains to science reporting, you've seen why such a technique is useful, but you've also seen 38 00:02:50,540 --> 00:02:52,210 the ways in which it is not useful. 39 00:02:52,760 --> 00:02:57,770 You now understand some of the real world issues that data scientists face when trying to transform 40 00:02:57,770 --> 00:03:01,300 data in such a way that can be easily communicated to others.