Again it is worth remarking that this model is somewhat simplistic. In particular, running back is the position where handcuffing is probably the most important and this model ignores it. To name one example, I would guess that the "true expectation" for any foursome that includes Corey Dillon and Laurence Maroney is a bit greater than the corresponding generic foursome including RB24 and RB33.

Also, I doubt there are too many leagues where each team only rosters four receivers. It would probably be more realistic to assume each team rosters six or seven. But seven receivers per team would mean 84 drafted receivers, and there are over 4.5 billion ways to choose seven from a group of 84. That's beyond the capabilities of my computer. Even bumping it up from four to five increases the number of computations required by a factor of about 30, and the program takes long enough to run as it is. We'll just have to live with four receivers and four running backs.

Finally, for the sake of completeness, here are the expected point values for each tight end pair, assuming you draft two, start one, and get two points for a waiver wire tight end.

      1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17  18  19  20  21  22  23
 1
 2  154
 3  132 123
 4  138 118 103
 5  134 123 106 104
 6  139 105 109  92 114
 7  133 121  98  96 100 102
 8  129 116  96  82  99  88  85
 9  128 111  96  81 105  90  82  82
10  136 129 109 106 111 103  90  98  91
11  132 119 101  93 103 100  96  89  80  95
12  135 109 104  80 107  87  85  83  73  86  80
13  133 111  95  78 107  81  91  83  75  82  82  68
14  131 119 103  95 103  93  93  89  76  79  80  78  73
15  128 110  96  82 103  90  82  83  70  70  70  68  71  71
16  131 110 100  85 103  86  82  78  71  75  70  65  63  58  66
17  133 108 105  85 112  94  95  89  81  93  86  72  74  81  77  70
18  132 106  99  74 103  86  80  79  71  75  72  62  58  65  60  50  66
19  135 111 100  80 107  88  86  84  76  86  81  73  65  76  72  67  72  65
20  133 106  96  78 103  87  90  80  74  82  81  78  62  76  71  61  68  57  68
21  129 105  97  81 100  89  86  86  71  75  77  71  64  71  65  62  71  62  68  66
22  133 110  98  81  98  83  82  80  73  72  75  69  61  67  63  55  68  52  66  55  60
23  130 110  98  83  98  84  82  78  73  71  77  63  58  67  64  56  72  53  68  62  57  48
24  135 107  99  87 103  83  83  82  75  72  76  64  61  66  66  55  75  53  68  62  62  48  42

A preview of Part III...

Now how do we put all this together? In the next article I will conduct a theoretical 12-team, 12-round draft where each team takes two quarterbacks, four running backs, four receivers, and two tight ends. Given those constraints, there are 207,900 ways a given team can order its picks (e.g. RB, RB, QB, WR, WR, RB, QB, TE, WR, WR, RB, TE would be one possibility. QB, RB, WR, WR, WR, RB, QB, TE, WR, TE, RB, RB would be another.)

I will look at the draft from each draft slot. So, if you're drafting in the #4 position, for example, and you opt to go QB, RB, WR, WR, WR, RB, QB, TE, WR, TE, RB, RB, what players will likely be available at the time you make your picks? Probably QB1, RB15, RB20, and so on. Or something like that. I'll use a current ADP list to make a reasonable estimate. Finally, I'll compute the expected points of the quarterbacks, running backs, receivers, and tight ends that you'd get using that particular order and add them up. Find the order with the highest overall point total and you've got yourself an "optimal" draft strategy for that slot.

Optimal is in quotes above because the resulting strategy will probably not be the optimal one for all the reasons mentioned in this article and the previous one (and a few more to be mentioned in the next one). This model is certainly too simplified to be able to recommend a truly optimal strategy in a real fantasy football league like the one you'll be drafting in this month.

So why do it? Mainly because it's fun; when my mind gets going down a path like this, I can't not take it to its logical conclusion. But we also might learn something. The model is naive, but I think it's realistic enough that if it gives us answers that are way out of line with our intuition, then they might actually be telling us something. If, for example, the model says that drafting a wide receiver with the first overall pick is a good idea (I doubt it will, but who knows?), then we'll have something to think about. If something like that happens, we can go in and investigate that particular case in detail and determine if we're getting that bizarre result because of weaknesses in the model or if we're getting that result because the model is showing us something we hadn't thought of before.