What’s More Likely? Part 1

This is the first in a multi-part series of posts where we look at whether a characteristic about a game has any relationship to which team will cover. In this post, we will explore whether the favorite or the underdog is more likely to cover the spread. To do this, we will work with the data set of college basketball games to determine whether the favorite or underdog is more likely to cover the spread.

First, some high level statistics. In the 22,215 games, the favorite covered 49% of the games, the underdog covered 48.9% of the games and 2.1% of games ended in a push (i.e., the outcome of the game matched the spread and neither team covered). So, neither the favorite or the underdog is more likely to cover the spread all else equal.

Is the Home Team More Likely to Cover the Spread?

No. In the 16,039 games the home team was favored they covered the spread in 8,037 games (50.1%). Conversely, the away underdog covered 8,002 spreads (49.9%). In the 5,713 games the home team was an underdog they covered the spread in 2,821 games (49.4%) and the away favorite covered 2,892 games (50.6%). In 463 games the spread resulted in a push.

So the home team is more likely to be favored but no more likely to cover than an away underdog. And the home underdog is no more likely to cover than the away favorite. So no pattern here…

Does the Size of the Spread Matter?

Another reasonable consideration is the size of the spread. Let’s see if a big favorite is more or less likely to cover the spread?

Putting the size of the spreads into 10 bins, we see no pattern.

A 1 point favorite has a 51% chance to cover the spread
A 16 point favorite has a 50.4% chance to cover the spread.

So no pattern here either….

Spread Range	Favorite Covers	Underdog Covers	Favorite Covers %
-1 to -1.5	813	778	51.0%
-2 to -2.5	1,045	1,050	49.9%
-3 to -3.5	976	994	49.6%
-4 to -5	1,404	1,401	50.0%
-5.5 to -6	905	858	51.3%
-6.5 to -7.5	1,202	1,214	49.8%
-8 to -9.5	1,249	1,262	49.7%
-10 to -12	1,092	1,132	49.1%
-12.5 to -15.5	1,079	1,054	50.6%
-16 or more	1,135	1,115	50.4%

There are many more factors that could contribute, but from the data available, we can conclude that neither playing at home, on the road, with a large spread or a small spread has any relationship to whether a team covers the spread.

You Might Want to Moneyline an Underdog

The team you want to bet on is 3.5 point underdogs. You have two choices, you can take the points and bet on the spread or you can take your underdog to win outright and bet on the moneyline. Typically taking the points will require you to wager $110 to win $100 (110/100) depending on the vig. The moneyline bet will give you odds, paying $150 for risk $100 (100/150).

What should you do? Take the moneyline bet every time.

Why? Because the expected value of the moneyline bet is significantly higher than the spread bet. Since we’ve already decided to bet on the underdog, what we need to know is what the percentage of the time that the 3.5 point spread will come into play, that is the underdog will lose the game by 3 points or fewer, resulting in the underdog covering the spread but losing the game. Then we need to determine whether the moneyline risk reward compensates us enough to offset the additional risk.

A few statistics from our basketball data set, specifically the 1,025 games with a 3.5 point spread. Side note – a 3.5 point spread is one of the most common type of spreads in college basketball.

The underdog won 409 games (39.9%)
The underdog covered the spread in 528 games (51.5%)
The underdog covered the spread but lost the game in 119 games (11.6%)

Using these facts, the expected outcome of each bet is calculated below.

Underdog Spread = -$1.74
- (.515*100)+(.484*-110)
Underdog Moneyline = -$.85
- (.393*150)+(.607*-100)

As you can see, the expected payout is for both bets is negative. Sports books wouldn’t be in business over the long run if they set positive payout scenarios. But more importantly, the moneyline expected payout is $.89 higher. While this doesn’t seem like much, it’s worth noting that card counting at Blackjack raises the payout per $100 wagered from -$1 loss to a $1 game or a difference of $2. This same strategy which takes no work increases your expected payout by 45% of the advantage card counting gets you. You may also note that the moneyline odds could be +155 or +160 for this line which would increase your payout.

A few notes

The spread and the moneyline, while related, do not always move in lockstep. For the above example, I could find instances where a 3.5 point spread was listed for moneyline odds at +145 to +165 and this plays an important part in the analysis with the greater the payout increasing the appeal of moneyline.
A different and less formulaic way to think about this problem is to think about the percentage of time you will lose a moneyline bet but would have won the spread bet. In this instance, we know these odds – 11.6%. You are risking around 50% more betting spreads instead of moneyline to win 11.6% more wagers.
This same logic holds for betting on a favorite with the moneyline in this example. However, analysis has shown that this rule is much less likely to hold at other odds. View this post for more details.
- Favorite Spread = -$8.25
  - (.484*100)+(.515*-110)
- Favorite Moneyline = -$6.11
  - (.393*-170)+(.607*100)

How Likely Is The Favorite to Win?

Your team is favored by 8 points. How likely does this make them to win?

In college basketball, there is a 77% chance a team favored by 8 points will win. In fact, each point in the spread increases the chance of winning by about 3.5%. A good rule of thumb is that 50%+(3.5 x spread) will get you within 1-2% of the chance of the favorite winning. However, this rule only applies up to about 11 point spreads at which point each additional point added to the spread is worth significantly less than 3.5% additional chance of winning.

Using 22,215 college basketball games played from November 2012 to January 2018, the actual data is displayed below in table format.

Spread	Games	Percentage of Games Won by Favorite
1	703	52.7%
1.5	913	54.7%
2	1,010	55.3%
2.5	1,117	56.9%
3	1,002	60.6%
3.5	1,025	60.0%
4	995	62.3%
4.5	999	66.4%
5	892	66.8%
5.5	960	71.1%
6	848	69.8%
6.5	860	70.3%
7	781	76.6%
7.5	804	77.1%
8	710	77.0%
8.5	678	80.3%
9	577	81.1%
9.5	599	84.8%
10	540	82.2%
10.5	499	84.9%
11	452	85.8%
11.5	416	89.6%
12	378	89.1%
12.5	399	83.9%
13	362	89.7%
13.5	330	90%
14	301	92.3%
14.5	279	92.4%
15	258	95.7%
15.5	242	96.6%
16	193	94.8%
16.5	191	95.2%
17	191	93.1%
17.5	128	96.8%
18	153	98.0%
18.5	123	95.1%
19	117	98.2%
19.5	101	99.0%
20 or more	1,089	98.5%

Here is the table data in chart form. As you can see, a relatively straight line until 20 points at which point the n-size per spread drops below 100 and the chance of winning stays relatively flat at 98%-99%.

View the table data in chart form.

Notes

The highest spread with an underdog winner was 25.5. Interestingly, 3 teams won with that spread.
Spreads above 20 are rare, accounting for less than 5% of all games with a spread.

How Accurate Are Vegas Lines?

I’ve heard it many times…Vegas knows! But I wondered. How good are the lines?

I was unhappy with what I could find on the internet so I decided to look into it. To start, I defined what I wanted to know. How accurate are Vegas sports betting lines? To be more specific, I wanted to know the average difference between the line Vegas set for a game and the actual outcome. For example, New England was 4.5 point favorites over Philadelphia in Super Bowl LII. Philadelphia won the game 41-33. So in this example Vegas was 12.5 points off on setting the spread (Philadelphia was expected to lose by 4.5 points and won by 8 points).

But is this typical? In Super Bowl LI New England was 3 point favorites over Atlanta and won the game by 6 points. So Vegas was off by only 3 points in this example and pretty accurately predicted the outcome of the game.

In this instance, we need to look at a larger sample of games. For this analysis I choose college basketball for a few reasons.

College basketball is a mainstream sport which means many people want to place bets on it which is one check on accuracy.
College basketball has a lot of games, which means larger populations to analyze. There are 256 NFL games a year (32 teams play 16 games) while in college basketball there may be 4,000 games in a season with a spread. Larger data sets allow for different types of analysis.

For purposes of analyzing the data, the metric Absolute Difference from Spread (ADS) was calculated for every game in the data set. For 22,215 college basketball games in the population, the median ADS was 7 and the mean ADS was 8.6. To state this another way, half of college basketball games are within 7 points of the spread and half are greater than 7 points.

View the data in chart form.

Is this good? We will explore this in a later post.

Data

The data used for this analysis is 22,215 college basketball games played between November 9, 2012 and January 31, 2018.
Absolute Difference from Spread (ADS) is the difference between the predicted outcome of the game and the actual outcome of the game measured using spread. For example, if Team A is favored by 15 and wins by 13, the ADS is 2. However, if Team A is favored by 5 and loses by 2 the ADS is 7.
The team with the largest ADS was Pittsburgh which on January 24, 2017 was 5 point underdogs to Louisville yet lost the game by 55 points. By the ADS measure, this was the most lopsided game in the last five years.