Interesting article on the (in)accuracy of race time predictors (Read 527 times)

juniordo1


    Just I know lots and lots of folks who train for the marathon with low mileage plans, though, for whom 1.15 might be accurate. 

    n=1: I ran my one and only marathon on a low milage plan. The 1.15 was very close (2:05Tight lippedx, 4:48Tight lippedx)

    2013 -Sub 2:00 for 1/2 marathon

    GC100k


      It could also be determined by plotting race pace vs log(d) on semi-log graph paper. The coefficient was specific for the person. 

       

      Similar idea but seems unnecessarily awkward because you end up with units and stuff.  Seems a lot easier to just put your times and distances in a spreadsheet and calculate an exponent in a minute.  I linearized it (take the log) and fit the exponent as the straight-line fit, but you can just right-click and have excel fit the exponent (which is exactly the same thing).  Once you have your exponent you can extrapolate to any distance you want.  And you can choose whatever races you want to put into your fitting - recent, prs, worst, whatever you want.

       

      If you want, you could set up a spreadsheet and estimate your times with 1.06 (~same performance level), 1.15 (average half to full), and your own n.  For example, let's say I run a 21 min 5k, a 45 min 10k, and 1:45 half.  Fitting that I get n = 1.12.  btw, no reason you couldn't put a bunch of 5ks and 10ks in there.  Starting with the 1:45 half, n = 1.06 would predict a 3:43 full, 1.12 (starting with the half or just extending the trend line) => 3:48, and 1.15 => 3:53.

       

      You could play around with the numbers using different starting points and different exponents to get a feel for the different predictions.  When I first started racing (late 90s), I took every race and every training course pr and extrapolated it do every other distance from 800m to marathon using n = 1.07.  I'm not much of a racer, I just did it out of curiosity.  No reason you need to rely on some online tool.

        I put my 7k and 10k into the calculator, it predicts my marathon is 3:43:14. If I use 5k and 7k, it gives 3:12:42, but if I just use 10K alone, it gives 3:23:12. Why 7 and 10K produces slower?

         

        Because your 5k PR is extremely soft--your 5k pace is the same as your 7k pace, which makes no sense. You haven't established enough of a baseline yet.

         

        Obviously your 5k does not represent your true fitness at that distance, so throw that out. If the 7k and 10k both represent your true fitness for those distances, then your paces drop off quickly as distance goes up.

         

        And who has a 7k PR??

        Runners run

        DoppleBock


          I have only run 2 or 3 serious half marathons close to my serious marathon - All were run full out

           

          1:25:57 / 2:56:10

          1:24:18 / 2:49:19

          1:24:03 / 2:48:10

           

          #2 and #3 were rough 1/2 Marathons with a slow down in the last couple miles

           

          Other times when I ran a 1/2M 10 miles @ MP and then ran the last 5k hard

           

          1:21:47 / 2:50:00 (Marathon a hot melt down at Grandmas)

          1:22:22 / 2:45:57

           

          One year I ran my marathon 1st and 1/2 later

          2:56:07 / 1:26:06

           

          I am a high mileage person (until this year - cutting way back) avg @ 5,000 a year for the last 5 or 6 years.  I do not have much raw speed, so my 1/2M to marathon time converts (If both are "Good races" to 2x +4 minutes.  So say on average 1:24 hald should convert to a 2:52 marathon.

           

          During some of these years, I have also ran a marathon for fun with my sister Between 4:29 - 4:50.

           

          Other years I can run full out marathon that are 20 minutes apart ~ just do to time of year and current fitness.  2012 I ran

          3:13 January so out of shape

          2:56 April moderate fitness

          1:26:06 June - drank 1 liter vodka night before

           

          How does one scrub data to actually compare the calculator for how it is meant to be used - Similar fitness level with enough endurance for it to translate.  To me the variation that you are seeing is the noise of dirty data.

           

           

           

          n=1: I ran my one and only marathon on a low milage plan. The 1.15 was very close (2:05Tight lippedx, 4:48Tight lippedx)

          Long dead ... But my stench lingers !

           

           

          wcrunner2


          Are we there, yet?

             

            Similar idea but seems unnecessarily awkward because you end up with units and stuff.  Seems a lot easier to just put your times and distances in a spreadsheet and calculate an exponent in a minute.  I linearized it (take the log) and fit the exponent as the straight-line fit, but you can just right-click and have excel fit the exponent (which is exactly the same thing).  Once you have your exponent you can extrapolate to any distance you want.  And you can choose whatever races you want to put into your fitting - recent, prs, worst, whatever you want.

             

            If you want, you could set up a spreadsheet and estimate your times with 1.06 (~same performance level), 1.15 (average half to full), and your own n.  For example, let's say I run a 21 min 5k, a 45 min 10k, and 1:45 half.  Fitting that I get n = 1.12.  btw, no reason you couldn't put a bunch of 5ks and 10ks in there.  Starting with the 1:45 half, n = 1.06 would predict a 3:43 full, 1.12 (starting with the half or just extending the trend line) => 3:48, and 1.15 => 3:53.

             

            You could play around with the numbers using different starting points and different exponents to get a feel for the different predictions.  When I first started racing (late 90s), I took every race and every training course pr and extrapolated it do every other distance from 800m to marathon using n = 1.07.  I'm not much of a racer, I just did it out of curiosity.  No reason you need to rely on some online tool.

            Nice idea but that wasn't available to me 10 years before Visicalc was in its first release.

             2024 Races:

                  03/09 - Livingston Oval Ultra 6-Hour, 22.88 miles

                  05/11 - D3 50K
                  05/25 - What the Duck 12-Hour

                  06/17 - 6 Days in the Dome 12-Hour.

             

             

                 

               

              Because your 5k PR is extremely soft--your 5k pace is the same as your 7k pace, which makes no sense. You haven't established enough of a baseline yet.

               

              Obviously your 5k does not represent your true fitness at that distance, so throw that out. If the 7k and 10k both represent your true fitness for those distances, then your paces drop off quickly as distance goes up.

               

              And who has a 7k PR??

               

              That makes sense.

               

              It was a HM relay (3 persons).

              5k - 20:56 (09/12), 7k - 28:40 (11/12), 10k trial - 43:08  (03/13), 42:05 (05/13), FM - 3:09:28 (05/13), HM - 1:28:20 (05/14), Failed 10K trial - 6:10/mi for 4mi (08/14), FM - 3:03 (09/14)

              GC100k


                Nice idea but that wasn't available to me 10 years before Visicalc was in its first release.

                 

                I get that.  Today, though, we too often rely on black-box models and tools when we could do the same thing in Excel, a program that most people use every day.  I'm not a real statistician, but I do teach applied stats in engineering sometimes.  I openly violate the recommendations of real statisticians that Excel not be used for teaching statistics.  You can do a lot in Excel, where you have all your data anyway, without the cost and learning curve of a separate stats program.  If you do stats a lot or need advanced methods, then a stats program will be much easier than the brute-force Excel method, but most of our students are better served with Excel.

                 

                That all being said, guys in the past were really clever before everyone had computers.  I'll show students methods that were developed using graph paper and log tables and nomographs and stuff.  Guys back then had to be inventive.

                 

                I'm not quite that old.  Our high school had the first (or maybe 2nd) generation Apple computers, we had the new TI calculators for Physics and Chem, and the PC came out when I was starting college.  So I was just barely too young for the pre-computer age.  My dad started as a computer programmer in 1959, programmed in machine language, and every time they bought a new computer he had to go to IBM in New York to learn how to program it.

                     No reason you need to rely on some online tool.

                   

                  But online tools are awesome!!!

                   

                  For fun, and since I have been playing around with the RunningAhead API, I built a quick online tool that allows you to analyze your race performance and calculate the coefficient using races stored in your log from 2010 to now.

                   

                  You can find it here: http://runningahead-api-test.appspot.com/

                   

                  For example  (I do not have many races...)  :

                   

                  It will look somewhat sketchy since it will say something like dougnd is requesting permission to view, and modify your workouts but the application only looks through your log for races, then sends them to you.  It does not save the data or the access_token and it modifies nothing.  You can see the server side code yourself: http://runningahead-api-test.appspot.com/src/helloworld.py (and yes, that is some terribly written code).

                  HermosaBoy


                     

                    But online tools are awesome!!!

                     

                     

                     

                    It will look somewhat sketchy since it will say something like dougnd is requesting permission to view, and modify your workouts but the application only looks through your log for races, then sends them to you.  It does not save the data or the access_token and it modifies nothing.  You can see the server side code yourself: http://runningahead-api-test.appspot.com/src/helloworld.py (and yes, that is some terribly written code).

                     

                    How far back does it go?

                    And you can quote me as saying I was mis-quoted. Groucho Marx

                     

                    Rob

                       

                      How far back does it go?

                       

                      Currently, it is hard coded in at Jan. 1 2010 (if I programmed it right).  Originally, I envisioned that being configurable, but this project was mostly just  a break from what I should be doing.

                      LedLincoln


                      not bad for mile 25

                         It will look somewhat sketchy since it will say something like dougnd is requesting permission to view, and modify your workouts but the application only looks through your log for races, then sends them to you.  It does not save the data or the access_token and it modifies nothing.  You can see the server side code yourself: http://runningahead-api-test.appspot.com/src/helloworld.py (and yes, that is some terribly written code).

                         

                        Error: Server Error

                        The server encountered an error and could not complete your request.

                        If the problem persists, please report your problem and mention this error message and the query that caused it.

                           Error: Server Error

                          LedLincoln,

                           

                          Hmmm, the traceback shows that it is dying when trying to parse data I get from RA.  I modified it to print out the JSON object it is trying to parse if parsing it fails.  So If you run it again now it will likely print out something like:

                           

                          {"activityName": "Run", "workoutID": 6, "workoutName": "Race", "raceStats": {"fieldSize": 814, "fieldPlacement": 327}, "course": {"id": "01938419384902385sdfkljalaj34j", "name": "2013 Goodwill Mud Run"}, "activityID": 10, "weather": {"conditions": ["overcast"], "temperature": {"value": 60, "unit": "F"}}, "details": {"duration": 3433.0, "distance": {"value": 3.5, "unit": "mi"}}, "time": "09:45:00", "date": "2013-04-14", "attributes": {"excludePR": true}, "id": "d0897asg98df7gas098gfas9df99s8df", "channel": "standard"}

                           

                          If you post that here, or pm it to me, I can try to figure out what is going on (if there is personal information or whatever, you can remove that bit, I suspect that random non-printable characters are being sent, perhaps in the Notes section).  Thanks for reporting it.  

                           

                          Doug

                            Worked for me, pretty cool.

                            Runners run


                            Why is it sideways?

                              Extremely cool. I had 37 races in that time, and they are all touching the line except for my marathon PR, which is just to the right of the line.

                               

                              Coeff of 1.076.

                                35 races and 1.076 for me, but if I take out the one outlier, Monkey, (which wasn't really a race) I get 1.0647.

                                Runners run