Great discussion, however I think that @ebeowulf17 is correct in that it's just a personal project, not anything that I will be trying to sell. So, I think I just over-reacted to the concern about having accurate resistors. I don't have much experience, so I can probably get by with less accurate resistors. Digikey is great, but I found that you have to order a certain dollar volume to get price/shipping down to reasonable values.
I don't often have enough of an order to make it worth it for Digikey.
For small orders it doesn't really work out well and I typically am better off going to ebay for china parts in small quantities with free shipping.

Thanks everyone, I feel like I learned a lot in the discusson but will probably try just some 1% resistors and see how stable it is and then maybe go down to 0.1% if I need to for stability and not try to get super precise/accurate right from the start.

 

First. This is spreadsheet work. You don’t need to waste people’s time with this.

Second.
In series accuracy will change additivly.
In parallel accuracy will change multiplicativly.

 

In series accuracy will change additivly.
In parallel accuracy will change multiplicativly.

Could you expand on what you mean by that?

 

First. This is spreadsheet work. You don’t need to waste people’s time with this.

I think that's a little harsh. It's only spreadsheet work once you understand the concepts well enough to build a meaningful, accurate spreadsheet. First you have to understand the concepts. Asking questions on the internet is one way to get a better understanding of the concepts. Admittedly, there's a good chance this could've been answered with google searches and reading existing documents instead of posting a new question, but either way you can't expect to build useful spreadsheets about something before you even understand how it works.

 

I think that's a little harsh. It's only spreadsheet work once you understand the concepts well enough to build a meaningful, accurate spreadsheet. First you have to understand the concepts. Asking questions on the internet is one way to get a better understanding of the concepts. Admittedly, there's a good chance this could've been answered with google searches and reading existing documents instead of posting a new question, but either way you can't expect to build useful spreadsheets about something before you even understand how it works.

How is it harsh. I take 1k 1% resistor. I calculate the lowest value. I calculate the hieghest value. Those are my two extreme cases.

For serial. I add two of my low numbers and I get X. Then i add two high values and I get Y. That is it. When I have two 1k 1% resistors in series I can expect to see an overall resistance between X and Y. All done in spreadsheet. Use google docs if you need free spreadsheet app and have internetz.

 

It has been many years since I did it, but testing revealed that, for the Philips resistors tested, the majority of a batch would differ from nominal by about the same amount and very few would have the opposite sign of error - e.g. nominally 1k 5% resistors might almost all be clustered close to 1020 ohms and only the occasional one would be less than 1000 ohms.

The thing I find most unsatisfactory about moderately high precision resistors is the relatively poor temperature coefficient of resistance. Low-cost 0.1% tolerance SM resistors typically have a tempco of ±25 ppm/°C, which is only "twice as good" as the Philips 1% tolerance resistors I used to use a lot. 10 ppm tempco ups the price considerably. However, I do admit that good SM resistors are much cheaper than equivalent-performance through-hole parts were.

I would much prefer to specify a 0.1% 10 ppm resistor than allow a twiddlable part into my design. I regard every trimpot as a defeat unless it is actually required to cope with something beyond my control. The pots are big and costly, the time it takes to adjust them is costly and they encourage maintenance bozos to leap to the the conclusion something is out of adjustment and therefore knobs should be turned. If you are going to tweak something, the system and the instruments you use must be up to the task.

 

How is it harsh. I take 1k 1% resistor. I calculate the lowest value. I calculate the hieghest value. Those are my two extreme cases.

For serial. I add two of my low numbers and I get X. Then i add two high values and I get Y. That is it. When I have two 1k 1% resistors in series I can expect to see an overall resistance between X and Y. All done in spreadsheet. Use google docs if you need free spreadsheet app and have internetz.

The spreadsheet idea isn't harsh. But saying that the thread starter is wasting people's time is harsh.

If someone doesn't already understand how errors combine with each other, then they'll need help understanding that. The spreadsheet doesn't tell them WHEN to add, WHEN to subtract, etc.

The thread starter didn't ask us to do the math on dozens of resistors for him. The thread starter didn't ask us to choose resistors for him. All the thread starter asked was:

I was wondering how the Accuracy of a resistor changes when you hook them up in parallel or series.
For example, if I have 2x 1% accuracy resistors and I hook them up in Series or Parallel, does the accuracy of the result double to 0.5%?

I think that is a VERY reasonable question. We're not all born knowing all this stuff. You have to start somewhere. When you have questions, you ask. Sometimes you ask a search engine or a textbook, sometimes you ask an internet forum. But a generic spreadsheet program doesn't answer the fundamental questions of how errors combine. The user has to tell the spreadsheet how to combine them, and then the spreadsheet can do the number crunching.

 

I think that's a little harsh. It's only spreadsheet work once you understand the concepts well enough to build a meaningful, accurate spreadsheet. First you have to understand the concepts. ...

Absolutely!

 

Here’s a table from a book I have in which the author spends 10 pages on resistor value distributions. He says resistors *tend* to have a Gaussian distribution within their tolerance, so combining them tends to result in better accuracy. This is just average values, not the mathematical extremes of the tolerance range. This illustrates what Wbahn said in post #17.

 

In series accuracy will change additivly.
In parallel accuracy will change multiplicativly.

The percent accuracy does not change whether in series or parallel.
If they are all 1% resistors then their serial or parallel worst-case accuracy is still 1% of the total resistance.

 

Here’s a table from a book I have in which the author spends 10 pages on resistor value distributions. He says resistors *tend* to have a Gaussian distribution within their tolerance, so combining them tends to result in better accuracy. This is just average values, not the mathematical extremes of the tolerance range. This illustrates what Wbahn said in post #17.

Does he give any relationship between the tolerance and the standard deviation of the distribution?

 

No he only gives two anecdotal sets of data. The focus was more on how to use the statistics to your advantage.

In one he quotes someone who measured 211 10k 1% metal film resistors, and found the standard deviation was 0.1%. Then he did his own test of 100 1k 1% metal film resistors and found the standard deviation was 0.21%, and that 99 out of the 100 were within 0.5% tolerance.

Source: Small Signal Audio Design by Douglas Self.

 

The problem that pops up in production of products is that while the values curve for an infinite number of samples may be fairly gaussian that does not mean that any particular batch of resistors will be at any specific value. So the way to avoid grief is to not depend on the values being so very exact. A design that can live with a bit more variation will not only be less challenging to produce with an adequate production yield, it will probably last longer in service as well. That is important to some applications.